fix(mcp): disable MCP_SKILLS feature flag — source not mirrored

Closes #856.

MCP servers that expose resources (e.g. RepoPrompt) failed to load
their tools in the open build with:

    Error fetching tools/commands/resources:
    fetchMcpSkillsForClient is not a function

Root cause: scripts/build.ts set MCP_SKILLS: true, which made
feature('MCP_SKILLS') evaluate to true at build time. The guards
around the dynamic skill discovery path therefore stayed live. The
underlying source file src/skills/mcpSkills.ts is not mirrored into
the open tree, so the bundler fell back to its generic missing-module
stub — which only exports `default` for require()-style imports, not
the named `fetchMcpSkillsForClient` binding. At runtime the require
returned an object without that property, and calling it threw.

`openclaude mcp doctor` reported RepoPrompt as healthy because doctor
does not exercise the skills-fetch path.

Fix: flip MCP_SKILLS to false and move it into the "Disabled: missing
source" group. With the flag off, every `if (feature('MCP_SKILLS'))`
guard becomes a no-op at build time, the require() branch is dead
code, and MCP servers with resources load normally via the existing
`Promise.resolve([])` fallbacks already present at each call site.

Also adds scripts/feature-flags-source-guard.test.ts to fail fast if
MCP_SKILLS (or any future flag in the same category) is re-enabled
without the corresponding source file being mirrored first.

Verification:
  - Test fails on main, passes with this fix
  - `bun run build` produces a bundle with no
    `missing-module-stub:../../skills/mcpSkills.js` reference
  - Full `bun test` — 1222 pass / 12 fail (same pre-existing 12 as
    main; new test adds the +1 pass)

.env.example

@@ -149,6 +149,23 @@ ANTHROPIC_API_KEY=sk-ant-your-key-here
 # Use a custom OpenAI-compatible endpoint (optional — defaults to api.openai.com)
 # OPENAI_BASE_URL=https://api.openai.com/v1
 
+# Fallback context window size (tokens) when the model is not found in the
+# built-in table (default: 128000). Increase this for models with larger
+# context windows (e.g. 200000 for Claude-sized contexts).
+# CLAUDE_CODE_OPENAI_FALLBACK_CONTEXT_WINDOW=128000
+
+# Per-model context window overrides as a JSON object.
+# Takes precedence over the built-in table, so you can register new or
+# custom models without patching source.
+# Example: CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS={"my-corp/llm-v3":262144,"gpt-4o-mini":128000}
+# CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS=
+
+# Per-model maximum output token overrides as a JSON object.
+# Use this alongside CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS when your model
+# supports a different output limit than what the built-in table specifies.
+# Example: CLAUDE_CODE_OPENAI_MAX_OUTPUT_TOKENS={"my-corp/llm-v3":8192}
+# CLAUDE_CODE_OPENAI_MAX_OUTPUT_TOKENS=
+
 
 # -----------------------------------------------------------------------------
 # Option 3: Google Gemini
@@ -267,6 +284,16 @@ ANTHROPIC_API_KEY=sk-ant-your-key-here
 # Disable "Co-authored-by" line in git commits made by OpenClaude
 # OPENCLAUDE_DISABLE_CO_AUTHORED_BY=1
 
+# Disable strict tool schema normalization for non-Gemini providers
+# Useful when MCP tools with complex optional params (e.g. list[dict])
+# trigger "Extra required key ... supplied" errors from OpenAI-compatible endpoints
+# OPENCLAUDE_DISABLE_STRICT_TOOLS=1
+
+# Disable hidden <system-reminder> messages injected into tool output
+# Suppresses the file-read cyber-risk reminder and the todo/task tool nudges
+# Useful for users who want full transparency over what the model sees
+# OPENCLAUDE_DISABLE_TOOL_REMINDERS=1
+
 # Custom timeout for API requests in milliseconds (default: varies)
 # API_TIMEOUT_MS=60000
 

.gitignore

@@ -11,3 +11,4 @@ CLAUDE.md
 package-lock.json
 /.claude
 coverage/
+agent.log

docs/hook-chains.md

@@ -0,0 +1,333 @@
+# Hook Chains (Self-Healing Agent Mesh MVP)
+
+Hook Chains provide an event-driven recovery layer for important workflow failures.
+When a matching hook event occurs, OpenClaude evaluates declarative rules and can dispatch remediation actions such as:
+
+- `spawn_fallback_agent`
+- `notify_team`
+- `warm_remote_capacity`
+
+## Disabled-By-Default Rollout
+
+> **Rollout recommendation:** keep Hook Chains disabled until you validate rules in your environment.
+>
+> - Set top-level config to `"enabled": false` initially.
+> - Enable per environment when ready.
+> - Dispatch is gated by `feature('HOOK_CHAINS')`.
+> - Env gate defaults to off unless `CLAUDE_CODE_ENABLE_HOOK_CHAINS=1` is set.
+
+This keeps existing workflows unchanged while you tune guard windows and action behavior.
+
+## Feature Overview
+
+Hook Chains are loaded from a deterministic config file and evaluated on dispatched hook events.
+
+MVP runtime trigger wiring:
+
+- `PostToolUseFailure` hooks dispatch Hook Chains with outcome `failed`.
+- `TaskCompleted` hooks dispatch Hook Chains with outcome:
+  - `success` when completion hooks did not block.
+  - `failed` when completion hooks returned blocking errors or prevented continuation.
+
+Default config path:
+
+- `.openclaude/hook-chains.json`
+
+Override path:
+
+- `CLAUDE_CODE_HOOK_CHAINS_CONFIG_PATH=/abs/or/relative/path/to/hook-chains.json`
+
+Global gate:
+
+- `feature('HOOK_CHAINS')` must be enabled in the build
+- `CLAUDE_CODE_ENABLE_HOOK_CHAINS=0|1` (defaults to disabled when unset)
+
+## Safety Guarantees
+
+The runtime is intentionally conservative:
+
+- **Depth guard:** chain dispatch is blocked when `chainDepth >= maxChainDepth`.
+- **Rule cooldown:** each rule can only re-fire after cooldown expires.
+- **Dedup window:** identical event/action combinations are suppressed for a window.
+- **Abort-safe behavior:** if the current signal is aborted, actions skip safely.
+- **Policy-aware remote warm:** `warm_remote_capacity` skips when remote sessions are policy denied.
+- **Bridge inactive no-op:** `warm_remote_capacity` safely skips when no active bridge handle exists.
+- **Missing team context safety:** `notify_team` skips with structured reason if no team context/team file is available.
+- **Fallback launcher safety:** `spawn_fallback_agent` fails with a structured reason when launch permissions/context are unavailable.
+
+## Configuration Schema Reference
+
+Top-level object:
+
+```json
+{
+  "version": 1,
+  "enabled": true,
+  "maxChainDepth": 2,
+  "defaultCooldownMs": 30000,
+  "defaultDedupWindowMs": 30000,
+  "rules": []
+}
+```
+
+### Top-Level Fields
+
+| Field | Type | Required | Notes |
+|---|---|---:|---|
+| `version` | `1` | No | Defaults to `1`. |
+| `enabled` | `boolean` | No | Global feature switch for this config file. |
+| `maxChainDepth` | `integer` | No | Global depth guard (default `2`, max `10`). |
+| `defaultCooldownMs` | `integer` | No | Default rule cooldown in ms (default `30000`). |
+| `defaultDedupWindowMs` | `integer` | No | Default action dedup window in ms (default `30000`). |
+| `rules` | `HookChainRule[]` | No | Defaults to `[]`. May be omitted or empty; when no rules are present, dispatch is a no-op and returns `enabled: false`. |
+
+> **Note:** An empty ruleset is valid and can be used to keep Hook Chains configured but effectively disabled until rules are added.
+### Rule Object (`HookChainRule`)
+
+```json
+{
+  "id": "task-failure-recovery",
+  "enabled": true,
+  "trigger": {
+    "event": "TaskCompleted",
+    "outcome": "failed"
+  },
+  "condition": {
+    "toolNames": ["Edit"],
+    "taskStatuses": ["failed"],
+    "errorIncludes": ["timeout", "permission denied"],
+    "eventFieldEquals": {
+      "meta.source": "scheduler"
+    }
+  },
+  "cooldownMs": 60000,
+  "dedupWindowMs": 30000,
+  "maxDepth": 2,
+  "actions": []
+}
+```
+
+| Field | Type | Required | Notes |
+|---|---|---:|---|
+| `id` | `string` | Yes | Stable identifier used in telemetry/guards. |
+| `enabled` | `boolean` | No | Per-rule switch. |
+| `trigger.event` | `HookEvent` | Yes | Event name to match. |
+| `trigger.outcome` | `"success"|"failed"|"timeout"|"unknown"` | No | Single outcome matcher. |
+| `trigger.outcomes` | `Outcome[]` | No | Multi-outcome matcher. Use either `outcome` or `outcomes`. |
+| `condition` | `object` | No | Optional extra matching constraints. |
+| `cooldownMs` | `integer` | No | Overrides global cooldown for this rule. |
+| `dedupWindowMs` | `integer` | No | Overrides global dedup for this rule. |
+| `maxDepth` | `integer` | No | Per-rule depth cap. |
+| `actions` | `HookChainAction[]` | Yes | One or more actions to execute in order. |
+
+### Condition Fields
+
+| Field | Type | Notes |
+|---|---|---|
+| `toolNames` | `string[]` | Matches `tool_name` / `toolName` in event payload. |
+| `taskStatuses` | `string[]` | Matches `task_status` / `taskStatus` / `status`. |
+| `errorIncludes` | `string[]` | Case-insensitive substring match against `error` / `reason` / `message`. |
+| `eventFieldEquals` | `Record<string, string\|number\|boolean>` | Dot-path equality against payload (example: `"meta.source": "scheduler"`). |
+
+### Actions
+
+#### `spawn_fallback_agent`
+
+```json
+{
+  "type": "spawn_fallback_agent",
+  "id": "fallback-1",
+  "enabled": true,
+  "dedupWindowMs": 30000,
+  "description": "Fallback recovery for failed task",
+  "promptTemplate": "Recover task ${TASK_SUBJECT}. Event=${EVENT_NAME}, outcome=${OUTCOME}, error=${ERROR}. Payload=${PAYLOAD_JSON}",
+  "agentType": "general-purpose",
+  "model": "sonnet"
+}
+```
+
+#### `notify_team`
+
+```json
+{
+  "type": "notify_team",
+  "id": "notify-ops",
+  "enabled": true,
+  "dedupWindowMs": 30000,
+  "teamName": "mesh-team",
+  "recipients": ["*"],
+  "summary": "Hook chain ${RULE_ID} fired",
+  "messageTemplate": "Event=${EVENT_NAME} outcome=${OUTCOME}\nTask=${TASK_ID}\nError=${ERROR}\nPayload=${PAYLOAD_JSON}"
+}
+```
+
+#### `warm_remote_capacity`
+
+```json
+{
+  "type": "warm_remote_capacity",
+  "id": "warm-bridge",
+  "enabled": true,
+  "dedupWindowMs": 60000,
+  "createDefaultEnvironmentIfMissing": false
+}
+```
+
+## Complete Example Configs
+
+### 1) Retry via Fallback Agent
+
+```json
+{
+  "version": 1,
+  "enabled": true,
+  "maxChainDepth": 2,
+  "defaultCooldownMs": 30000,
+  "defaultDedupWindowMs": 30000,
+  "rules": [
+    {
+      "id": "retry-task-via-fallback",
+      "trigger": {
+        "event": "TaskCompleted",
+        "outcome": "failed"
+      },
+      "cooldownMs": 60000,
+      "actions": [
+        {
+          "type": "spawn_fallback_agent",
+          "id": "spawn-retry-agent",
+          "description": "Retry failed task with fallback agent",
+          "promptTemplate": "A task failed. Recover it safely.\nTask=${TASK_SUBJECT}\nDescription=${TASK_DESCRIPTION}\nError=${ERROR}\nPayload=${PAYLOAD_JSON}",
+          "agentType": "general-purpose",
+          "model": "sonnet"
+        }
+      ]
+    }
+  ]
+}
+```
+
+### 2) Notify Only
+
+```json
+{
+  "version": 1,
+  "enabled": true,
+  "maxChainDepth": 2,
+  "defaultCooldownMs": 30000,
+  "defaultDedupWindowMs": 30000,
+  "rules": [
+    {
+      "id": "notify-on-tool-failure",
+      "trigger": {
+        "event": "PostToolUseFailure",
+        "outcome": "failed"
+      },
+      "condition": {
+        "toolNames": ["Edit", "Write", "Bash"]
+      },
+      "actions": [
+        {
+          "type": "notify_team",
+          "id": "notify-team-failure",
+          "recipients": ["*"],
+          "summary": "Tool failure detected",
+          "messageTemplate": "Tool failure detected.\nEvent=${EVENT_NAME} outcome=${OUTCOME}\nError=${ERROR}\nPayload=${PAYLOAD_JSON}"
+        }
+      ]
+    }
+  ]
+}
+```
+
+### 3) Combined Fallback + Notify + Bridge Warm
+
+```json
+{
+  "version": 1,
+  "enabled": true,
+  "maxChainDepth": 2,
+  "defaultCooldownMs": 45000,
+  "defaultDedupWindowMs": 30000,
+  "rules": [
+    {
+      "id": "full-recovery-chain",
+      "trigger": {
+        "event": "TaskCompleted",
+        "outcomes": ["failed", "timeout"]
+      },
+      "condition": {
+        "errorIncludes": ["timeout", "capacity", "connection"]
+      },
+      "cooldownMs": 90000,
+      "actions": [
+        {
+          "type": "spawn_fallback_agent",
+          "id": "fallback-agent",
+          "description": "Recover failed task execution",
+          "promptTemplate": "Recover failed task and produce a concise fix summary.\nTask=${TASK_SUBJECT}\nError=${ERROR}\nPayload=${PAYLOAD_JSON}"
+        },
+        {
+          "type": "notify_team",
+          "id": "notify-team",
+          "recipients": ["*"],
+          "summary": "Recovery chain triggered",
+          "messageTemplate": "Recovery chain ${RULE_ID} fired.\nOutcome=${OUTCOME}\nTask=${TASK_SUBJECT}\nError=${ERROR}"
+        },
+        {
+          "type": "warm_remote_capacity",
+          "id": "warm-capacity",
+          "createDefaultEnvironmentIfMissing": false
+        }
+      ]
+    }
+  ]
+}
+```
+
+## Template Variables
+
+The following placeholders are supported by `promptTemplate`, `summary`, and `messageTemplate`:
+
+- `${EVENT_NAME}`
+- `${OUTCOME}`
+- `${RULE_ID}`
+- `${TASK_SUBJECT}`
+- `${TASK_DESCRIPTION}`
+- `${TASK_ID}`
+- `${ERROR}`
+- `${PAYLOAD_JSON}`
+
+## Troubleshooting
+
+### Rule never triggers
+
+- Verify `trigger.event` and `trigger.outcome`/`trigger.outcomes` exactly match dispatched event data.
+- Check `condition` filters (especially `toolNames` and `eventFieldEquals` dot-path keys).
+- Confirm the config file is valid JSON and schema-valid.
+
+### Actions show as skipped
+
+Common skip reasons:
+
+- `action disabled`
+- `rule cooldown active ...`
+- `dedup window active ...`
+- `max chain depth reached ...`
+- `No team context is available ...`
+- `Team file not found ...`
+- `Remote sessions are blocked by policy`
+- `Bridge is not active; warm_remote_capacity is a safe no-op`
+- `No fallback agent launcher is registered in runtime context`
+
+### Config changes not reflected
+
+- Loader uses memoization by file mtime/size.
+- Ensure your editor writes the file fully and updates mtime.
+- If needed, force reload from the caller side with `forceReloadConfig: true`.
+
+### Existing workflows changed unexpectedly
+
+- Set `"enabled": false` at top-level.
+- Or globally disable with `CLAUDE_CODE_ENABLE_HOOK_CHAINS=0`.
+- Re-enable gradually after validating one rule at a time.

scripts/build.ts

@@ -34,6 +34,7 @@ const featureFlags: Record<string, boolean> = {
   WEB_BROWSER_TOOL: false,        // Built-in browser automation (source not mirrored)
   CHICAGO_MCP: false,             // Computer-use MCP (native Swift modules stubbed)
   COWORKER_TYPE_TELEMETRY: false, // Telemetry for agent/coworker type classification
+  MCP_SKILLS: false,              // Dynamic MCP skill discovery (src/skills/mcpSkills.ts not mirrored; enabling this causes "fetchMcpSkillsForClient is not a function" when MCP servers with resources connect — see #856)
 
   // ── Enabled: upstream defaults ──────────────────────────────────────
   COORDINATOR_MODE: true,             // Multi-agent coordinator with worker delegation
@@ -56,7 +57,6 @@ const featureFlags: Record<string, boolean> = {
   EXTRACT_MEMORIES: true,             // Auto-extract durable memories from conversations
   FORK_SUBAGENT: true,                // Implicit context-forking when omitting subagent_type
   VERIFICATION_AGENT: true,           // Built-in read-only agent for test/verification
-  MCP_SKILLS: true,                   // Discover skills dynamically from MCP server resources
   PROMPT_CACHE_BREAK_DETECTION: true, // Detect & log unexpected prompt cache invalidations
   HOOK_PROMPTS: true,                 // Allow tools to request interactive user prompts
 }

scripts/feature-flags-source-guard.test.ts

@@ -0,0 +1,47 @@
+import { existsSync, readFileSync } from 'fs'
+import { join } from 'path'
+import { expect, test } from 'bun:test'
+
+// Regression guard for #856. Several build feature flags require source files
+// that are not mirrored into the open build. When such a flag is set to `true`
+// without the source present, the bundler falls back to a missing-module stub
+// that only exports `default`, which causes runtime errors like
+// `fetchMcpSkillsForClient is not a function` when downstream code reaches
+// through the `require()` to a named export.
+//
+// This test fails fast at test-time if someone re-enables one of these flags
+// without first mirroring the corresponding source file.
+
+const BUILD_SCRIPT = join(import.meta.dir, 'build.ts')
+const REPO_ROOT = join(import.meta.dir, '..')
+
+type FlagGuard = {
+  flag: string
+  source: string // path relative to repo root
+}
+
+const FLAG_REQUIRES_SOURCE: FlagGuard[] = [
+  { flag: 'MCP_SKILLS', source: 'src/skills/mcpSkills.ts' },
+]
+
+test('build feature flags are not enabled without their source files', () => {
+  const buildScript = readFileSync(BUILD_SCRIPT, 'utf-8')
+
+  for (const { flag, source } of FLAG_REQUIRES_SOURCE) {
+    const enabledRe = new RegExp(`^\\s*${flag}\\s*:\\s*true\\b`, 'm')
+    const isEnabled = enabledRe.test(buildScript)
+    const sourceExists = existsSync(join(REPO_ROOT, source))
+
+    if (isEnabled && !sourceExists) {
+      throw new Error(
+        `Feature flag ${flag} is enabled in scripts/build.ts, but its required source file "${source}" does not exist. ` +
+          `Enabling this flag without the source will cause runtime errors (missing named exports from the missing-module stub). ` +
+          `Either mirror the source file or set ${flag}: false.`,
+      )
+    }
+
+    // When the source IS present, the flag can be either true or false; either
+    // is fine. We only care about the "enabled but missing" combination.
+    expect(true).toBe(true)
+  }
+})

src/Tool.ts

@@ -249,6 +249,11 @@ export type ToolUseContext = {
   /** When true, canUseTool must always be called even when hooks auto-approve.
    *  Used by speculation for overlay file path rewriting. */
   requireCanUseTool?: boolean
+  /**
+   * Optional callback used by hook-chain fallback actions that launch
+   * AgentTool from hook runtime paths.
+   */
+  hookChainsCanUseTool?: CanUseToolFn
   messages: Message[]
   fileReadingLimits?: {
     maxTokens?: number

src/__tests__/bugfixes.test.ts

@@ -169,6 +169,14 @@ describe('Web search result count improvements', () => {
 
     expect(content).toMatch(/max_uses:\s*15/)
   })
+
+  test('codex web search path guarantees a non-empty result body', async () => {
+    const content = await file(
+      'tools/WebSearchTool/WebSearchTool.ts',
+    ).text()
+
+    expect(content).toContain("results.push('No results found.')")
+  })
 })
 
 // ---------------------------------------------------------------------------

src/components/StartupScreen.test.ts

@@ -0,0 +1,158 @@
+import { afterEach, beforeEach, describe, expect, test } from 'bun:test'
+import { detectProvider } from './StartupScreen.js'
+
+const ENV_KEYS = [
+  'CLAUDE_CODE_USE_OPENAI',
+  'CLAUDE_CODE_USE_GEMINI',
+  'CLAUDE_CODE_USE_GITHUB',
+  'CLAUDE_CODE_USE_BEDROCK',
+  'CLAUDE_CODE_USE_VERTEX',
+  'CLAUDE_CODE_USE_MISTRAL',
+  'OPENAI_BASE_URL',
+  'OPENAI_API_KEY',
+  'OPENAI_MODEL',
+  'GEMINI_MODEL',
+  'MISTRAL_MODEL',
+  'ANTHROPIC_MODEL',
+  'NVIDIA_NIM',
+  'MINIMAX_API_KEY',
+]
+
+const originalEnv: Record<string, string | undefined> = {}
+
+beforeEach(() => {
+  for (const key of ENV_KEYS) {
+    originalEnv[key] = process.env[key]
+    delete process.env[key]
+  }
+})
+
+afterEach(() => {
+  for (const key of ENV_KEYS) {
+    if (originalEnv[key] === undefined) {
+      delete process.env[key]
+    } else {
+      process.env[key] = originalEnv[key]
+    }
+  }
+})
+
+function setupOpenAIMode(baseUrl: string, model: string): void {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = baseUrl
+  process.env.OPENAI_MODEL = model
+  process.env.OPENAI_API_KEY = 'test-key'
+}
+
+// --- Issue #855: aggregator URL must win over vendor-prefixed model name ---
+
+describe('detectProvider — aggregator URL authoritative over model-name substring (#855)', () => {
+  test('OpenRouter + deepseek/deepseek-chat labels as OpenRouter', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'deepseek/deepseek-chat')
+    expect(detectProvider().name).toBe('OpenRouter')
+  })
+
+  test('OpenRouter + moonshotai/kimi-k2 labels as OpenRouter', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'moonshotai/kimi-k2')
+    expect(detectProvider().name).toBe('OpenRouter')
+  })
+
+  test('OpenRouter + mistralai/mistral-large labels as OpenRouter', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'mistralai/mistral-large')
+    expect(detectProvider().name).toBe('OpenRouter')
+  })
+
+  test('OpenRouter + meta-llama/llama-3.3 labels as OpenRouter', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'meta-llama/llama-3.3-70b-instruct')
+    expect(detectProvider().name).toBe('OpenRouter')
+  })
+
+  test('Together + deepseek-ai/DeepSeek-V3 labels as Together AI', () => {
+    setupOpenAIMode('https://api.together.xyz/v1', 'deepseek-ai/DeepSeek-V3')
+    expect(detectProvider().name).toBe('Together AI')
+  })
+
+  test('Together + meta-llama/Llama-3.3 labels as Together AI', () => {
+    setupOpenAIMode('https://api.together.xyz/v1', 'meta-llama/Llama-3.3-70B-Instruct-Turbo')
+    expect(detectProvider().name).toBe('Together AI')
+  })
+
+  test('Groq + deepseek-r1-distill-llama-70b labels as Groq', () => {
+    setupOpenAIMode('https://api.groq.com/openai/v1', 'deepseek-r1-distill-llama-70b')
+    expect(detectProvider().name).toBe('Groq')
+  })
+
+  test('Groq + llama-3.3-70b-versatile labels as Groq', () => {
+    setupOpenAIMode('https://api.groq.com/openai/v1', 'llama-3.3-70b-versatile')
+    expect(detectProvider().name).toBe('Groq')
+  })
+
+  test('Azure + any deepseek deployment labels as Azure OpenAI', () => {
+    setupOpenAIMode('https://my-resource.openai.azure.com/', 'deepseek-chat')
+    expect(detectProvider().name).toBe('Azure OpenAI')
+  })
+})
+
+// --- Direct vendor endpoints still label correctly (regression) ---
+
+describe('detectProvider — direct vendor endpoints', () => {
+  test('api.deepseek.com labels as DeepSeek', () => {
+    setupOpenAIMode('https://api.deepseek.com/v1', 'deepseek-chat')
+    expect(detectProvider().name).toBe('DeepSeek')
+  })
+
+  test('api.moonshot.cn labels as Moonshot (Kimi)', () => {
+    setupOpenAIMode('https://api.moonshot.cn/v1', 'moonshot-v1-8k')
+    expect(detectProvider().name).toBe('Moonshot (Kimi)')
+  })
+
+  test('api.mistral.ai labels as Mistral', () => {
+    setupOpenAIMode('https://api.mistral.ai/v1', 'mistral-large-latest')
+    expect(detectProvider().name).toBe('Mistral')
+  })
+
+  test('default OpenAI URL + gpt-4o labels as OpenAI', () => {
+    setupOpenAIMode('https://api.openai.com/v1', 'gpt-4o')
+    expect(detectProvider().name).toBe('OpenAI')
+  })
+})
+
+// --- rawModel fallback for generic/custom endpoints ---
+
+describe('detectProvider — rawModel fallback when URL is generic', () => {
+  test('custom proxy + deepseek-chat falls back to DeepSeek', () => {
+    setupOpenAIMode('https://my-proxy.internal/v1', 'deepseek-chat')
+    expect(detectProvider().name).toBe('DeepSeek')
+  })
+
+  test('custom proxy + kimi-k2 falls back to Moonshot (Kimi)', () => {
+    setupOpenAIMode('https://my-proxy.internal/v1', 'kimi-k2-instruct')
+    expect(detectProvider().name).toBe('Moonshot (Kimi)')
+  })
+
+  test('custom proxy + llama-3.3 falls back to Meta Llama', () => {
+    setupOpenAIMode('https://my-proxy.internal/v1', 'llama-3.3-70b')
+    expect(detectProvider().name).toBe('Meta Llama')
+  })
+
+  test('custom proxy + mistral-large falls back to Mistral', () => {
+    setupOpenAIMode('https://my-proxy.internal/v1', 'mistral-large-latest')
+    expect(detectProvider().name).toBe('Mistral')
+  })
+})
+
+// --- Explicit env flags win over URL heuristics ---
+
+describe('detectProvider — explicit dedicated-provider env flags', () => {
+  test('NVIDIA_NIM=1 overrides aggregator URL', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'some-nim-model')
+    process.env.NVIDIA_NIM = '1'
+    expect(detectProvider().name).toBe('NVIDIA NIM')
+  })
+
+  test('MINIMAX_API_KEY overrides aggregator URL', () => {
+    setupOpenAIMode('https://openrouter.ai/api/v1', 'any-model')
+    process.env.MINIMAX_API_KEY = 'test-key'
+    expect(detectProvider().name).toBe('MiniMax')
+  })
+})

src/components/StartupScreen.ts

@@ -83,7 +83,7 @@ const LOGO_CLAUDE = [
 
 // ─── Provider detection ───────────────────────────────────────────────────────
 
-function detectProvider(): { name: string; model: string; baseUrl: string; isLocal: boolean } {
+export function detectProvider(): { name: string; model: string; baseUrl: string; isLocal: boolean } {
   const useGemini = process.env.CLAUDE_CODE_USE_GEMINI === '1' || process.env.CLAUDE_CODE_USE_GEMINI === 'true'
   const useGithub = process.env.CLAUDE_CODE_USE_GITHUB === '1' || process.env.CLAUDE_CODE_USE_GITHUB === 'true'
   const useOpenAI = process.env.CLAUDE_CODE_USE_OPENAI === '1' || process.env.CLAUDE_CODE_USE_OPENAI === 'true'
@@ -117,30 +117,34 @@ function detectProvider(): { name: string; model: string; baseUrl: string; isLoc
     const baseUrl = resolvedRequest.baseUrl
     const isLocal = isLocalProviderUrl(baseUrl)
     let name = 'OpenAI'
-    if (/nvidia/i.test(baseUrl) || /nvidia/i.test(rawModel) || process.env.NVIDIA_NIM)
+    // Explicit dedicated-provider env flags win.
-      name = 'NVIDIA NIM'
+    if (process.env.NVIDIA_NIM) name = 'NVIDIA NIM'
-    else if (/minimax/i.test(baseUrl) || /minimax/i.test(rawModel) || process.env.MINIMAX_API_KEY)
+    else if (process.env.MINIMAX_API_KEY) name = 'MiniMax'
-      name = 'MiniMax'
+    else if (
-    else if (resolvedRequest.transport === 'codex_responses' || baseUrl.includes('chatgpt.com/backend-api/codex'))
+      resolvedRequest.transport === 'codex_responses' ||
+      baseUrl.includes('chatgpt.com/backend-api/codex')
+    )
       name = 'Codex'
-    else if (/moonshot/i.test(baseUrl) || /kimi/i.test(rawModel))
+    // Base URL is authoritative — must precede rawModel checks so aggregators
-      name = 'Moonshot (Kimi)'
+    // (OpenRouter/Together/Groq) aren't mislabelled as DeepSeek/Kimi/etc.
-    else if (/deepseek/i.test(baseUrl) || /deepseek/i.test(rawModel))
+    // when routed to models whose IDs contain a vendor prefix. See issue #855.
-      name = 'DeepSeek'
+    else if (/openrouter/i.test(baseUrl)) name = 'OpenRouter'
-    else if (/openrouter/i.test(baseUrl))
+    else if (/together/i.test(baseUrl)) name = 'Together AI'
-      name = 'OpenRouter'
+    else if (/groq/i.test(baseUrl)) name = 'Groq'
-    else if (/together/i.test(baseUrl))
+    else if (/azure/i.test(baseUrl)) name = 'Azure OpenAI'
-      name = 'Together AI'
+    else if (/nvidia/i.test(baseUrl)) name = 'NVIDIA NIM'
-    else if (/groq/i.test(baseUrl))
+    else if (/minimax/i.test(baseUrl)) name = 'MiniMax'
-      name = 'Groq'
+    else if (/moonshot/i.test(baseUrl)) name = 'Moonshot (Kimi)'
-    else if (/mistral/i.test(baseUrl) || /mistral/i.test(rawModel))
+    else if (/deepseek/i.test(baseUrl)) name = 'DeepSeek'
-      name = 'Mistral'
+    else if (/mistral/i.test(baseUrl)) name = 'Mistral'
-    else if (/azure/i.test(baseUrl))
+    // rawModel fallback — fires only when base URL is generic/custom.
-      name = 'Azure OpenAI'
+    else if (/nvidia/i.test(rawModel)) name = 'NVIDIA NIM'
-    else if (/llama/i.test(rawModel))
+    else if (/minimax/i.test(rawModel)) name = 'MiniMax'
-      name = 'Meta Llama'
+    else if (/kimi/i.test(rawModel)) name = 'Moonshot (Kimi)'
-    else if (isLocal)
+    else if (/deepseek/i.test(rawModel)) name = 'DeepSeek'
-      name = getLocalOpenAICompatibleProviderLabel(baseUrl)
+    else if (/mistral/i.test(rawModel)) name = 'Mistral'
+    else if (/llama/i.test(rawModel)) name = 'Meta Llama'
+    else if (isLocal) name = getLocalOpenAICompatibleProviderLabel(baseUrl)
     
     // Resolve model alias to actual model name + reasoning effort
     let displayModel = resolvedRequest.resolvedModel

src/hooks/usePasteHandler.test.ts

@@ -1,5 +1,8 @@
 import { expect, test } from 'bun:test'
-import { supportsClipboardImageFallback } from './usePasteHandler.ts'
+import {
+  shouldHandleInputAsPaste,
+  supportsClipboardImageFallback,
+} from './usePasteHandler.ts'
 
 test('supports clipboard image fallback on Windows', () => {
   expect(supportsClipboardImageFallback('windows')).toBe(true)
@@ -20,3 +23,42 @@ test('does not support clipboard image fallback on WSL', () => {
 test('does not support clipboard image fallback on unknown platforms', () => {
   expect(supportsClipboardImageFallback('unknown')).toBe(false)
 })
+
+test('does not treat a bracketed paste as pending when no paste handlers are provided', () => {
+  expect(
+    shouldHandleInputAsPaste({
+      hasTextPasteHandler: false,
+      hasImagePasteHandler: false,
+      inputLength: 'kimi-k2.5'.length,
+      pastePending: false,
+      hasImageFilePath: false,
+      isFromPaste: true,
+    }),
+  ).toBe(false)
+})
+
+test('treats bracketed text paste as pending when a text paste handler exists', () => {
+  expect(
+    shouldHandleInputAsPaste({
+      hasTextPasteHandler: true,
+      hasImagePasteHandler: false,
+      inputLength: 'kimi-k2.5'.length,
+      pastePending: false,
+      hasImageFilePath: false,
+      isFromPaste: true,
+    }),
+  ).toBe(true)
+})
+
+test('treats image path paste as pending when only an image handler exists', () => {
+  expect(
+    shouldHandleInputAsPaste({
+      hasTextPasteHandler: false,
+      hasImagePasteHandler: true,
+      inputLength: 'C:\\Users\\jat\\image.png'.length,
+      pastePending: false,
+      hasImageFilePath: true,
+      isFromPaste: false,
+    }),
+  ).toBe(true)
+})

src/hooks/usePasteHandler.ts

@@ -35,6 +35,24 @@ type PasteHandlerProps = {
   ) => void
 }
 
+export function shouldHandleInputAsPaste(options: {
+  hasTextPasteHandler: boolean
+  hasImagePasteHandler: boolean
+  inputLength: number
+  pastePending: boolean
+  hasImageFilePath: boolean
+  isFromPaste: boolean
+}): boolean {
+  return (
+    (options.hasTextPasteHandler &&
+      (options.inputLength > PASTE_THRESHOLD ||
+        options.pastePending ||
+        options.hasImageFilePath ||
+        options.isFromPaste)) ||
+    (options.hasImagePasteHandler && options.hasImageFilePath)
+  )
+}
+
 export function usePasteHandler({
   onPaste,
   onInput,
@@ -236,11 +254,6 @@ export function usePasteHandler({
     // The keypress parser sets isPasted=true for content within bracketed paste.
     const isFromPaste = event.keypress.isPasted
 
-    // If this is pasted content, set isPasting state for UI feedback
-    if (isFromPaste) {
-      setIsPasting(true)
-    }
-
     // Handle large pastes (>PASTE_THRESHOLD chars)
     // Usually we get one or two input characters at a time. If we
     // get more than the threshold, the user has probably pasted.
@@ -268,6 +281,7 @@ export function usePasteHandler({
       canFallbackToClipboardImage &&
       onImagePaste
     ) {
+      setIsPasting(true)
       checkClipboardForImage()
       // Reset isPasting since there's no text content to process
       setIsPasting(false)
@@ -275,14 +289,17 @@ export function usePasteHandler({
     }
 
     // Check if we should handle as paste (from bracketed paste, large input, or continuation)
-    const shouldHandleAsPaste =
+    const shouldHandleAsPaste = shouldHandleInputAsPaste({
-      onPaste &&
+      hasTextPasteHandler: Boolean(onPaste),
-      (input.length > PASTE_THRESHOLD ||
+      hasImagePasteHandler: Boolean(onImagePaste),
-        pastePendingRef.current ||
+      inputLength: input.length,
-        hasImageFilePath ||
+      pastePending: pastePendingRef.current,
-        isFromPaste)
+      hasImageFilePath,
+      isFromPaste,
+    })
 
     if (shouldHandleAsPaste) {
+      setIsPasting(true)
       pastePendingRef.current = true
       setPasteState(({ chunks, timeoutId }) => {
         return {

src/services/api/codexShim.test.ts

@@ -8,6 +8,7 @@ import {
   convertCodexResponseToAnthropicMessage,
   convertToolsToResponsesTools,
 } from './codexShim.js'
+import { __test as webSearchToolTest } from '../../tools/WebSearchTool/WebSearchTool.js'
 
 const tempDirs: string[] = []
 const originalEnv = {
@@ -609,6 +610,164 @@ describe('Codex request translation', () => {
     ])
   })
 
+  test('recovers Codex web search text and sources from sparse completed response', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      {
+        output: [
+          {
+            type: 'web_search_call',
+            sources: [
+              {
+                title: 'OpenClaude repo',
+                url: 'https://github.com/example/openclaude',
+              },
+            ],
+          },
+          {
+            type: 'message',
+            role: 'assistant',
+            content: [
+              {
+                type: 'text',
+                text: 'OpenClaude is available on GitHub.',
+                sources: [
+                  {
+                    title: 'Docs',
+                    url: 'https://docs.example.com/openclaude',
+                  },
+                ],
+              },
+            ],
+          },
+        ],
+      },
+      'OpenClaude GitHub 2026',
+      0.42,
+    )
+
+    expect(output.results).toEqual([
+      'OpenClaude is available on GitHub.',
+      {
+        tool_use_id: 'codex-web-search',
+        content: [
+          {
+            title: 'OpenClaude repo',
+            url: 'https://github.com/example/openclaude',
+          },
+          {
+            title: 'Docs',
+            url: 'https://docs.example.com/openclaude',
+          },
+        ],
+      },
+    ])
+  })
+
+  test('falls back to a non-empty Codex web search result message', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      { output: [] },
+      'OpenClaude GitHub 2026',
+      0.11,
+    )
+
+    expect(output.results).toEqual(['No results found.'])
+  })
+
+  test('surfaces Codex web search failure reason with a message', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      {
+        output: [
+          {
+            type: 'web_search_call',
+            status: 'failed',
+            error: { message: 'upstream search provider rate-limited' },
+          },
+        ],
+      },
+      'OpenClaude GitHub 2026',
+      0.05,
+    )
+
+    expect(output.results).toEqual([
+      'Web search failed: upstream search provider rate-limited',
+    ])
+  })
+
+  test('surfaces Codex web search failure reason nested under action.error', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      {
+        output: [
+          {
+            type: 'web_search_call',
+            status: 'failed',
+            action: { error: { message: 'query blocked' } },
+          },
+        ],
+      },
+      'OpenClaude GitHub 2026',
+      0.05,
+    )
+
+    expect(output.results).toEqual(['Web search failed: query blocked'])
+  })
+
+  test('handles Codex web search failure with no reason attached', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      {
+        output: [
+          {
+            type: 'web_search_call',
+            status: 'failed',
+          },
+        ],
+      },
+      'OpenClaude GitHub 2026',
+      0.05,
+    )
+
+    expect(output.results).toEqual(['Web search failed.'])
+  })
+
+  test('a failure item does not suppress sources from a later message item', () => {
+    const output = webSearchToolTest.makeOutputFromCodexWebSearchResponse(
+      {
+        output: [
+          {
+            type: 'web_search_call',
+            status: 'failed',
+            error: { message: 'partial outage' },
+          },
+          {
+            type: 'message',
+            role: 'assistant',
+            content: [
+              {
+                type: 'output_text',
+                text: 'Partial results below.',
+                sources: [
+                  { title: 'Docs', url: 'https://docs.example.com/openclaude' },
+                ],
+              },
+            ],
+          },
+        ],
+      },
+      'OpenClaude GitHub 2026',
+      0.05,
+    )
+
+    expect(output.results).toEqual([
+      'Web search failed: partial outage',
+      'Partial results below.',
+      {
+        tool_use_id: 'codex-web-search',
+        content: [
+          { title: 'Docs', url: 'https://docs.example.com/openclaude' },
+        ],
+      },
+    ])
+  })
+
   test('translates Codex SSE text stream into Anthropic events', async () => {
     const responseText = [
       'event: response.output_item.added',

src/services/api/openaiShim.test.ts

@@ -3343,6 +3343,139 @@ test('Moonshot: uses max_tokens (not max_completion_tokens) and strips store', a
   expect(requestBody?.store).toBeUndefined()
 })
 
+test('Moonshot: echoes reasoning_content on assistant tool-call messages', async () => {
+  // Regression for: "API Error: 400 {"error":{"message":"thinking is enabled
+  // but reasoning_content is missing in assistant tool call message at index
+  // N"}}" when the agent sends a prior-turn assistant response back to Kimi.
+  // The thinking block captured from the inbound response must round-trip
+  // as reasoning_content on the outgoing echoed assistant message.
+  process.env.OPENAI_BASE_URL = 'https://api.moonshot.ai/v1'
+  process.env.OPENAI_API_KEY = 'sk-moonshot-test'
+
+  let requestBody: Record<string, unknown> | undefined
+  globalThis.fetch = (async (_input, init) => {
+    requestBody = JSON.parse(String(init?.body))
+    return new Response(
+      JSON.stringify({
+        id: 'chatcmpl-1',
+        model: 'kimi-k2.6',
+        choices: [
+          { message: { role: 'assistant', content: 'ok' }, finish_reason: 'stop' },
+        ],
+        usage: { prompt_tokens: 3, completion_tokens: 1, total_tokens: 4 },
+      }),
+      { headers: { 'Content-Type': 'application/json' } },
+    )
+  }) as FetchType
+
+  const client = createOpenAIShimClient({}) as OpenAIShimClient
+  await client.beta.messages.create({
+    model: 'kimi-k2.6',
+    system: 'you are kimi',
+    messages: [
+      { role: 'user', content: 'check the logs' },
+      {
+        role: 'assistant',
+        content: [
+          {
+            type: 'thinking',
+            thinking: 'Need to inspect logs via Bash; running a cat.',
+          },
+          { type: 'text', text: "I'll inspect the logs." },
+          {
+            type: 'tool_use',
+            id: 'call_bash_1',
+            name: 'Bash',
+            input: { command: 'cat /tmp/app.log' },
+          },
+        ],
+      },
+      {
+        role: 'user',
+        content: [
+          {
+            type: 'tool_result',
+            tool_use_id: 'call_bash_1',
+            content: 'log line 1\nlog line 2',
+          },
+        ],
+      },
+    ],
+    max_tokens: 256,
+    stream: false,
+  })
+
+  const messages = requestBody?.messages as Array<Record<string, unknown>>
+  const assistantWithToolCall = messages.find(
+    m => m.role === 'assistant' && Array.isArray(m.tool_calls),
+  )
+  expect(assistantWithToolCall).toBeDefined()
+  expect(assistantWithToolCall?.reasoning_content).toBe(
+    'Need to inspect logs via Bash; running a cat.',
+  )
+})
+
+test('non-Moonshot providers do NOT receive reasoning_content on assistant messages', async () => {
+  // Guard: only Moonshot opts in. DeepSeek/OpenRouter/etc. receive the
+  // outgoing assistant message without reasoning_content to avoid
+  // unknown-field rejections from strict servers.
+  process.env.OPENAI_BASE_URL = 'https://api.deepseek.com/v1'
+  process.env.OPENAI_API_KEY = 'sk-deepseek'
+
+  let requestBody: Record<string, unknown> | undefined
+  globalThis.fetch = (async (_input, init) => {
+    requestBody = JSON.parse(String(init?.body))
+    return new Response(
+      JSON.stringify({
+        id: 'chatcmpl-1',
+        model: 'deepseek-chat',
+        choices: [
+          { message: { role: 'assistant', content: 'ok' }, finish_reason: 'stop' },
+        ],
+        usage: { prompt_tokens: 3, completion_tokens: 1, total_tokens: 4 },
+      }),
+      { headers: { 'Content-Type': 'application/json' } },
+    )
+  }) as FetchType
+
+  const client = createOpenAIShimClient({}) as OpenAIShimClient
+  await client.beta.messages.create({
+    model: 'deepseek-chat',
+    system: 'test',
+    messages: [
+      { role: 'user', content: 'hi' },
+      {
+        role: 'assistant',
+        content: [
+          { type: 'thinking', thinking: 'thought' },
+          { type: 'text', text: 'hello' },
+          {
+            type: 'tool_use',
+            id: 'call_1',
+            name: 'Bash',
+            input: { command: 'ls' },
+          },
+        ],
+      },
+      {
+        role: 'user',
+        content: [
+          { type: 'tool_result', tool_use_id: 'call_1', content: 'files' },
+        ],
+      },
+    ],
+    max_tokens: 32,
+    stream: false,
+  })
+
+  const messages = requestBody?.messages as Array<Record<string, unknown>>
+  const assistantWithToolCall = messages.find(
+    m => m.role === 'assistant' && Array.isArray(m.tool_calls),
+  )
+  expect(assistantWithToolCall).toBeDefined()
+  expect(assistantWithToolCall?.reasoning_content).toBeUndefined()
+})
+
 test('Moonshot: cn host is also detected', async () => {
   process.env.OPENAI_BASE_URL = 'https://api.moonshot.cn/v1'
   process.env.OPENAI_API_KEY = 'sk-moonshot-test'

src/services/api/openaiShim.ts

@@ -67,6 +67,8 @@ import {
   normalizeToolArguments,
   hasToolFieldMapping,
 } from './toolArgumentNormalization.js'
+import { logApiCallStart, logApiCallEnd } from '../../utils/requestLogging.js'
+import { createStreamState, processStreamChunk, getStreamStats } from '../../utils/streamingOptimizer.js'
 
 type SecretValueSource = Partial<{
   OPENAI_API_KEY: string
@@ -216,6 +218,14 @@ interface OpenAIMessage {
   }>
   tool_call_id?: string
   name?: string
+  /**
+   * Per-assistant-message chain-of-thought, attached when echoing an
+   * assistant message back to providers that require it (notably Moonshot:
+   * "thinking is enabled but reasoning_content is missing in assistant
+   * tool call message at index N" 400). Derived from the Anthropic thinking
+   * block captured when the original response was translated.
+   */
+  reasoning_content?: string
 }
 
 interface OpenAITool {
@@ -383,7 +393,9 @@ function convertMessages(
     content?: unknown
   }>,
   system: unknown,
+  options?: { preserveReasoningContent?: boolean },
 ): OpenAIMessage[] {
+  const preserveReasoningContent = options?.preserveReasoningContent === true
   const result: OpenAIMessage[] = []
   const knownToolCallIds = new Set<string>()
 
@@ -486,6 +498,21 @@ function convertMessages(
           })(),
         }
 
+        // Providers that validate reasoning continuity (Moonshot: "thinking
+        // is enabled but reasoning_content is missing in assistant tool call
+        // message at index N" 400) need the original chain-of-thought echoed
+        // back on each assistant message that carries a tool_call. We kept
+        // the thinking block on the Anthropic side; re-attach it here as the
+        // `reasoning_content` field on the outgoing OpenAI-shaped message.
+        // Gated per-provider because other endpoints either ignore the field
+        // (harmless) or strict-reject unknown fields (harmful).
+        if (preserveReasoningContent) {
+          const thinkingText = (thinkingBlock as { thinking?: string } | undefined)?.thinking
+          if (typeof thinkingText === 'string' && thinkingText.trim().length > 0) {
+            assistantMsg.reasoning_content = thinkingText
+          }
+        }
+
         if (toolUses.length > 0) {
           const mappedToolCalls = toolUses
             .map(
@@ -857,6 +884,7 @@ async function* openaiStreamToAnthropic(
   let lastStopReason: 'tool_use' | 'max_tokens' | 'end_turn' | null = null
   let hasEmittedFinalUsage = false
   let hasProcessedFinishReason = false
+  const streamState = createStreamState()
 
   // Emit message_start
   yield {
@@ -1020,6 +1048,7 @@ async function* openaiStreamToAnthropic(
               delta: { type: 'text_delta', text: visible },
             }
           }
+          processStreamChunk(streamState, delta.content)
         }
 
         // Tool calls
@@ -1039,6 +1068,7 @@ async function* openaiStreamToAnthropic(
               const toolBlockIndex = contentBlockIndex
               const initialArguments = tc.function.arguments ?? ''
               const normalizeAtStop = hasToolFieldMapping(tc.function.name)
+              processStreamChunk(streamState, tc.function.arguments ?? '')
               activeToolCalls.set(tc.index, {
                 id: tc.id,
                 name: tc.function.name,
@@ -1236,6 +1266,20 @@ async function* openaiStreamToAnthropic(
     reader.releaseLock()
   }
 
+  const stats = getStreamStats(streamState)
+  if (stats.totalChunks > 0) {
+    logForDebugging(
+      JSON.stringify({
+        type: 'stream_stats',
+        model,
+        total_chunks: stats.totalChunks,
+        first_token_ms: stats.firstTokenMs,
+        duration_ms: stats.durationMs,
+      }),
+      { level: 'debug' },
+    )
+  }
+
   yield { type: 'message_stop' }
 }
 
@@ -1441,7 +1485,12 @@ class OpenAIShimMessages {
       }>,
       request.resolvedModel,
     )
-    const openaiMessages = convertMessages(compressedMessages, params.system)
+    const openaiMessages = convertMessages(compressedMessages, params.system, {
+      // Moonshot requires every assistant tool-call message to carry
+      // reasoning_content when its thinking feature is active. Echo it back
+      // from the thinking block we captured on the inbound response.
+      preserveReasoningContent: isMoonshotBaseUrl(request.baseUrl),
+    })
 
     const body: Record<string, unknown> = {
       model: request.resolvedModel,
@@ -1715,6 +1764,12 @@ class OpenAIShimMessages {
     }
 
     let response: Response | undefined
+    const provider = request.baseUrl.includes('nvidia') ? 'nvidia-nim'
+      : request.baseUrl.includes('minimax') ? 'minimax'
+      : request.baseUrl.includes('localhost:11434') || request.baseUrl.includes('localhost:11435') ? 'ollama'
+      : request.baseUrl.includes('anthropic') ? 'anthropic'
+      : 'openai'
+    const { correlationId, startTime } = logApiCallStart(provider, request.resolvedModel)
     for (let attempt = 0; attempt < maxAttempts; attempt++) {
       try {
         response = await fetchWithProxyRetry(
@@ -1752,6 +1807,20 @@ class OpenAIShimMessages {
       }
 
       if (response.ok) {
+        let tokensIn = 0
+        let tokensOut = 0
+        // Skip clone() for streaming responses - it blocks until full body is received,
+        // defeating the purpose of streaming. Usage data is already sent via
+        // stream_options: { include_usage: true } and can be extracted from the stream.
+        if (!params.stream) {
+          try {
+            const clone = response.clone()
+            const data = await clone.json()
+            tokensIn = data.usage?.prompt_tokens ?? 0
+            tokensOut = data.usage?.completion_tokens ?? 0
+          } catch { /* ignore */ }
+        }
+        logApiCallEnd(correlationId, startTime, request.resolvedModel, 'success', tokensIn, tokensOut, false)
         return response
       }
 

src/services/tokenEstimation.ts

@@ -223,6 +223,49 @@ export function bytesPerTokenForFileType(fileExtension: string): number {
   }
 }
 
+/**
+ * Tokenizer ratio by model family.
+ * Different models have different encodings.
+ */
+export interface ModelTokenizerConfig {
+  modelFamily: string
+  bytesPerToken: number
+  supportsJson: boolean
+  supportsCode: boolean
+}
+
+export const MODEL_TOKENIZER_CONFIGS: ModelTokenizerConfig[] = [
+  { modelFamily: 'claude', bytesPerToken: 3.5, supportsJson: true, supportsCode: true },
+  { modelFamily: 'gpt-4', bytesPerToken: 4, supportsJson: true, supportsCode: true },
+  { modelFamily: 'gpt-3.5', bytesPerToken: 4, supportsJson: true, supportsCode: true },
+  { modelFamily: 'gemini', bytesPerToken: 3.5, supportsJson: true, supportsCode: true },
+  { modelFamily: 'llama', bytesPerToken: 3.8, supportsJson: true, supportsCode: true },
+  { modelFamily: 'deepseek', bytesPerToken: 3.5, supportsJson: true, supportsCode: true },
+  { modelFamily: 'minimax', bytesPerToken: 3.2, supportsJson: true, supportsCode: true },
+]
+
+/**
+ * Get tokenizer config for a model.
+ */
+export function getTokenizerConfig(model: string): ModelTokenizerConfig {
+  const lower = model.toLowerCase()
+  
+  for (const config of MODEL_TOKENIZER_CONFIGS) {
+    if (lower.includes(config.modelFamily)) {
+      return config
+    }
+  }
+  
+  return { modelFamily: 'unknown', bytesPerToken: 4, supportsJson: true, supportsCode: true }
+}
+
+/**
+ * Get bytes-per-token ratio for a model.
+ */
+export function getBytesPerTokenForModel(model: string): number {
+  return getTokenizerConfig(model).bytesPerToken
+}
+
 /**
  * Like {@link roughTokenCountEstimation} but uses a more accurate
  * bytes-per-token ratio when the file type is known.
@@ -241,6 +284,106 @@ export function roughTokenCountEstimationForFileType(
   )
 }
 
+/**
+ * Content type classification for compression ratio.
+ */
+export type ContentType = 
+  | 'json' | 'code' | 'prose' | 'technical' 
+  | 'list' | 'table' | 'mixed'
+
+/**
+ * Compression ratio by content type.
+ * Measured empirically - denser content = lower ratio.
+ */
+export const COMPRESSION_RATIOS: Record<ContentType, { min: number; max: number; typical: number }> = {
+  json: { min: 1.5, max: 2.5, typical: 2 },
+  code: { min: 3, max: 4.5, typical: 3.5 },
+  prose: { min: 3.5, max: 4.5, typical: 4 },
+  technical: { min: 2.5, max: 3.5, typical: 3 },
+  list: { min: 2, max: 3, typical: 2.5 },
+  table: { min: 1.8, max: 2.8, typical: 2.2 },
+  mixed: { min: 3, max: 4, typical: 3.5 },
+}
+
+/**
+ * Detect content type from content.
+ */
+export function detectContentType(content: string): ContentType {
+  const trimmed = content.trim()
+  
+  // JSON
+  if ((trimmed.startsWith('{') && trimmed.endsWith('}')) || 
+      (trimmed.startsWith('[') && trimmed.endsWith(']'))) {
+    try {
+      JSON.parse(trimmed)
+      return 'json'
+    } catch { /* not valid json */ }
+  }
+  
+  // Table (tabs or consistent delimiters)
+  const lines = trimmed.split('\n')
+  if (lines.length > 2) {
+    const hasTabs = lines[0].includes('\t')
+    const hasCommas = lines[0].includes(',')
+    if (hasTabs || hasCommas) {
+      const consistent = lines.slice(1).every(l => l.includes('\t') || l.includes(','))
+      if (consistent) return 'table'
+    }
+  }
+  
+  // List
+  if (/^[\d\-\*\•]/.test(trimmed) || /^[\d\-\*\•]/.test(lines[0])) {
+    return 'list'
+  }
+  
+  // Code (high density of special chars)
+  const codeChars = (content.match(/[{}()\[\];=]/g) || []).length
+  const codeRatio = codeChars / content.length
+  if (codeRatio > 0.05) return 'code'
+  
+  // Technical (has numbers and units)
+  if (/\d+\s*(px|em|rem|%|ms|s|kb|mb|gb)/i.test(content)) {
+    return 'technical'
+  }
+  
+  // Prose (default - natural language)
+  return 'prose'
+}
+
+/**
+ * Get compression ratio for content.
+ */
+export function getCompressionRatio(content: string, type?: ContentType): { ratio: number; min: number; max: number } {
+  const detectedType = type ?? detectContentType(content)
+  const { min, max, typical } = COMPRESSION_RATIOS[detectedType]
+  
+  // Adjust based on actual content length
+  // Shorter content = higher variance
+  const lengthBonus = content.length < 100 ? 0.5 : 0
+  
+  return {
+    ratio: typical,
+    min: min + lengthBonus,
+    max: max + lengthBonus,
+  }
+}
+
+/**
+ * Estimate tokens with confidence bounds.
+ */
+export function estimateWithBounds(
+  content: string,
+  type?: ContentType,
+): { estimate: number; min: number; max: number } {
+  const { ratio, min: minRatio, max: maxRatio } = getCompressionRatio(content, type)
+  
+  const estimate = roughTokenCountEstimation(content, ratio)
+  const min = roughTokenCountEstimation(content, maxRatio)
+  const max = roughTokenCountEstimation(content, minRatio)
+  
+  return { estimate, min, max }
+}
+
 /**
  * Estimates token count for a Message object by extracting and analyzing its text content.
  * This provides a more reliable estimate than getTokenUsage for messages that may have been compacted.

src/services/tokenModelCompression.test.ts

@@ -0,0 +1,100 @@
+import { describe, expect, it } from 'bun:test'
+import {
+  getTokenizerConfig,
+  getBytesPerTokenForModel,
+  detectContentType,
+  getCompressionRatio,
+  estimateWithBounds,
+} from './tokenEstimation.js'
+
+describe('Model Tokenizers', () => {
+  describe('getTokenizerConfig', () => {
+    it('returns config for claude models', () => {
+      const config = getTokenizerConfig('claude-sonnet-4-5-20250514')
+      expect(config.modelFamily).toBe('claude')
+      expect(config.bytesPerToken).toBe(3.5)
+    })
+
+    it('returns config for gpt models', () => {
+      const config = getTokenizerConfig('gpt-4')
+      expect(config.modelFamily).toBe('gpt-4')
+      expect(config.bytesPerToken).toBe(4)
+    })
+
+    it('returns default for unknown models', () => {
+      const config = getTokenizerConfig('unknown-model')
+      expect(config.modelFamily).toBe('unknown')
+      expect(config.bytesPerToken).toBe(4)
+    })
+  })
+
+  describe('getBytesPerTokenForModel', () => {
+    it('returns bytes per token for model', () => {
+      expect(getBytesPerTokenForModel('claude-opus-3-5-20250214')).toBe(3.5)
+      expect(getBytesPerTokenForModel('gpt-4o')).toBe(4)
+      expect(getBytesPerTokenForModel('deepseek-chat')).toBe(3.5)
+      expect(getBytesPerTokenForModel('minimax-M2.7')).toBe(3.2)
+    })
+  })
+})
+
+describe('Content Type Detection', () => {
+  describe('detectContentType', () => {
+    it('detects JSON', () => {
+      expect(detectContentType('{"key": "value"}')).toBe('json')
+      expect(detectContentType('[1, 2, 3]')).toBe('json')
+    })
+
+    it('detects code', () => {
+      expect(detectContentType('function test() { return 1 + 2; }')).toBe('code')
+      expect(detectContentType('const x = () => {}')).toBe('code')
+    })
+
+    it('detects prose', () => {
+      expect(detectContentType('This is a natural language response.')).toBe('prose')
+      expect(detectContentType('Hello world how are you?')).toBe('prose')
+    })
+
+    it('detects code-like technical', () => {
+      // Has both code chars and technical - higher code char ratio wins
+      expect(detectContentType('margin: 10px; padding: 5px;')).toBe('code')
+    })
+
+    it('detects list', () => {
+      expect(detectContentType('- item 1\n- item 2')).toBe('list')
+      expect(detectContentType('1. first\n2. second')).toBe('list')
+    })
+
+    it('detects prose by default', () => {
+      // Single column with newlines = prose
+      expect(detectContentType('a b c\n1 2 3')).toBe('prose')
+    })
+  })
+})
+
+describe('Compression Ratio', () => {
+  describe('getCompressionRatio', () => {
+    it('returns appropriate ratios', () => {
+      expect(getCompressionRatio('{"a":1}').ratio).toBe(2)
+      expect(getCompressionRatio('code here {} []').ratio).toBe(3.5)
+      expect(getCompressionRatio('Hello world').ratio).toBe(4)
+    })
+  })
+
+  describe('estimateWithBounds', () => {
+    it('returns estimate with bounds', () => {
+      const result = estimateWithBounds('Hello world')
+      
+      expect(result.min).toBeLessThanOrEqual(result.estimate)
+      expect(result.max).toBeGreaterThanOrEqual(result.estimate)
+      expect(result.min).toBeLessThan(result.max)
+    })
+
+    it('handles JSON with tighter bounds', () => {
+      const result = estimateWithBounds('{"key": "value"}')
+      
+      // JSON has smaller ratio range
+      expect(result.max).toBeLessThan(10)
+    })
+  })
+})

src/services/tools/toolExecution.ts

@@ -1241,6 +1241,7 @@ async function checkPermissionsAndCallTool(
       {
         ...toolUseContext,
         toolUseId: toolUseID,
+        hookChainsCanUseTool: canUseTool,
         userModified: permissionDecision.userModified ?? false,
       },
       canUseTool,
@@ -1729,19 +1730,29 @@ async function checkPermissionsAndCallTool(
     const hookMessages: MessageUpdateLazy<
       AttachmentMessage | ProgressMessage<HookProgress>
     >[] = []
-    for await (const hookResult of runPostToolUseFailureHooks(
+    const hookChainsContext = toolUseContext as ToolUseContext & {
-      toolUseContext,
+      hookChainsCanUseTool?: CanUseToolFn
-      tool,
+    }
-      toolUseID,
+    hookChainsContext.hookChainsCanUseTool = canUseTool
-      messageId,
+    try {
-      processedInput,
+      for await (const hookResult of runPostToolUseFailureHooks(
-      content,
+        toolUseContext,
-      isInterrupt,
+        tool,
-      requestId,
+        toolUseID,
-      mcpServerType,
+        messageId,
-      mcpServerBaseUrl,
+        processedInput,
-    )) {
+        content,
-      hookMessages.push(hookResult)
+        isInterrupt,
+        requestId,
+        mcpServerType,
+        mcpServerBaseUrl,
+      )) {
+        hookMessages.push(hookResult)
+      }
+    } finally {
+      if (hookChainsContext.hookChainsCanUseTool === canUseTool) {
+        delete hookChainsContext.hookChainsCanUseTool
+      }
     }
 
     return [

src/services/tools/toolHooks.ts

@@ -284,6 +284,7 @@ export async function* runPostToolUseFailureHooks<Input extends AnyObject>(
       isInterrupt,
       permissionMode,
       toolUseContext.abortController.signal,
+      undefined,
     )) {
       try {
         // Check if we were aborted during hook execution

src/tools/FileReadTool/FileReadTool.ts

@@ -733,6 +733,9 @@ export const CYBER_RISK_MITIGATION_REMINDER =
 const MITIGATION_EXEMPT_MODELS = new Set(['claude-opus-4-6'])
 
 function shouldIncludeFileReadMitigation(): boolean {
+  if (isEnvTruthy(process.env.OPENCLAUDE_DISABLE_TOOL_REMINDERS)) {
+    return false
+  }
   const shortName = getCanonicalName(getMainLoopModel())
   return !MITIGATION_EXEMPT_MODELS.has(shortName)
 }

src/tools/WebFetchTool/applyPromptFallback.test.ts

@@ -0,0 +1,87 @@
+import { afterEach, beforeEach, expect, mock, test } from 'bun:test'
+
+// Mock the Anthropic-API-side before importing the module under test, so
+// queryHaiku resolves into whatever the individual test wants (slow, failing,
+// or successful). We preserve every other export from claude.js so unrelated
+// transitive imports still work.
+const haikuMock = mock()
+
+beforeEach(async () => {
+  haikuMock.mockReset()
+  const actual = await import('../../services/api/claude.js')
+  mock.module('../../services/api/claude.js', () => ({
+    ...actual,
+    queryHaiku: haikuMock,
+  }))
+})
+
+afterEach(() => {
+  mock.restore()
+})
+
+async function runApply(markdown = 'Hello world.', signal?: AbortSignal): Promise<string> {
+  const nonce = `${Date.now()}-${Math.random()}`
+  const { applyPromptToMarkdown } =
+    await import(`./utils.js?ts=${nonce}`)
+  const ctrl = new AbortController()
+  return applyPromptToMarkdown(
+    'summarize',
+    markdown,
+    signal ?? ctrl.signal,
+    false,
+    false,
+  )
+}
+
+test('returns raw truncated markdown when queryHaiku throws', async () => {
+  haikuMock.mockImplementation(async () => {
+    throw new Error('MiniMax rejected the model name')
+  })
+
+  const output = await runApply('Gitlawb homepage content.')
+  expect(output).toContain('[Secondary-model summarization unavailable')
+  expect(output).toContain('Gitlawb homepage content.')
+})
+
+test('returns raw truncated markdown when queryHaiku simulates a timeout', async () => {
+  // Simulating raceWithTimeout's rejection path directly — we can't actually
+  // wait 45s in a test. The error shape matches what raceWithTimeout produces.
+  haikuMock.mockImplementation(async () => {
+    const err = new Error('Secondary-model summarization timed out after 45000ms')
+    ;(err as NodeJS.ErrnoException).code = 'SECONDARY_MODEL_TIMEOUT'
+    throw err
+  })
+
+  const output = await runApply('Slow provider content.')
+  expect(output).toContain('[Secondary-model summarization unavailable')
+  expect(output).toContain('Slow provider content.')
+})
+
+test('returns the model response when queryHaiku succeeds', async () => {
+  haikuMock.mockImplementation(async () => ({
+    message: {
+      content: [{ type: 'text', text: 'This page is about GitLawb, an AI legal platform.' }],
+    },
+  }))
+
+  const output = await runApply('some page content')
+  expect(output).toBe('This page is about GitLawb, an AI legal platform.')
+})
+
+test('returns fallback when queryHaiku resolves with empty content', async () => {
+  haikuMock.mockImplementation(async () => ({ message: { content: [] } }))
+
+  const output = await runApply('some page content')
+  expect(output).toContain('[Secondary-model summarization unavailable')
+  expect(output).toContain('some page content')
+})
+
+test('propagates AbortError from the caller signal', async () => {
+  const ctrl = new AbortController()
+  haikuMock.mockImplementation(async () => {
+    ctrl.abort()
+    return new Promise(() => {})
+  })
+
+  await expect(runApply('content', ctrl.signal)).rejects.toThrow()
+})

src/tools/WebFetchTool/domainCheck.test.ts

@@ -20,8 +20,11 @@ afterEach(() => {
 describe('checkDomainBlocklist', () => {
   test('returns allowed without API call in OpenAI mode', async () => {
     process.env.CLAUDE_CODE_USE_OPENAI = '1'
+    const actual = await import('../../utils/model/providers.js')
     mock.module('../../utils/model/providers.js', () => ({
+      ...actual,
       getAPIProvider: () => 'openai',
+      isFirstPartyAnthropicBaseUrl: () => false,
     }))
     const getSpy = mock(() =>
       Promise.resolve({ status: 200, data: { can_fetch: true } }),
@@ -37,8 +40,11 @@ describe('checkDomainBlocklist', () => {
 
   test('returns allowed without API call in Gemini mode', async () => {
     process.env.CLAUDE_CODE_USE_GEMINI = '1'
+    const actual = await import('../../utils/model/providers.js')
     mock.module('../../utils/model/providers.js', () => ({
+      ...actual,
       getAPIProvider: () => 'gemini',
+      isFirstPartyAnthropicBaseUrl: () => false,
     }))
     const getSpy = mock(() =>
       Promise.resolve({ status: 200, data: { can_fetch: true } }),
@@ -57,8 +63,11 @@ describe('checkDomainBlocklist', () => {
     delete process.env.CLAUDE_CODE_USE_GEMINI
     delete process.env.CLAUDE_CODE_USE_GITHUB
 
+    const actual = await import('../../utils/model/providers.js')
     mock.module('../../utils/model/providers.js', () => ({
+      ...actual,
       getAPIProvider: () => 'firstParty',
+      isFirstPartyAnthropicBaseUrl: () => true,
     }))
     const getSpy = mock(() =>
       Promise.resolve({ status: 200, data: { can_fetch: true } }),

src/tools/WebFetchTool/utils.ts

@@ -275,20 +275,76 @@ export async function getWithPermittedRedirects(
   if (depth > MAX_REDIRECTS) {
     throw new Error(`Too many redirects (exceeded ${MAX_REDIRECTS})`)
   }
+
+  const axiosConfig = {
+    signal,
+    timeout: FETCH_TIMEOUT_MS,
+    maxRedirects: 0,
+    responseType: 'arraybuffer' as const,
+    maxContentLength: MAX_HTTP_CONTENT_LENGTH,
+    lookup: ssrfGuardedLookup,
+    headers: {
+      Accept: 'text/markdown, text/html, */*',
+      'User-Agent': getWebFetchUserAgent(),
+    },
+  }
+
   try {
-    return await axios.get(url, {
+    return await axios.get(url, axiosConfig)
-      signal,
-      timeout: FETCH_TIMEOUT_MS,
-      maxRedirects: 0,
-      responseType: 'arraybuffer',
-      maxContentLength: MAX_HTTP_CONTENT_LENGTH,
-      lookup: ssrfGuardedLookup,
-      headers: {
-        Accept: 'text/markdown, text/html, */*',
-        'User-Agent': getWebFetchUserAgent(),
-      },
-    })
   } catch (error) {
+    // Try native fetch as a fallback for timeout / network errors
+    // (Bun/Node bundled contexts occasionally hang with axios + custom lookup.)
+    const isTimeoutLike =
+      axios.isAxiosError(error) &&
+      (!error.response &&
+        (error.code === 'ECONNABORTED' ||
+          error.code === 'ETIMEDOUT' ||
+          error.message?.toLowerCase().includes('timeout')))
+    if (isTimeoutLike && !signal.aborted) {
+      try {
+        const fetchResponse = await fetch(url, {
+          signal,
+          redirect: 'manual',
+          headers: axiosConfig.headers,
+        })
+        // Handle redirects manually
+        if ([301, 302, 307, 308].includes(fetchResponse.status)) {
+          const redirectLocation = fetchResponse.headers.get('location')
+          if (!redirectLocation) {
+            throw new Error('Redirect missing Location header')
+          }
+          const redirectUrl = new URL(redirectLocation, url).toString()
+          if (redirectChecker(url, redirectUrl)) {
+            return getWithPermittedRedirects(
+              redirectUrl,
+              signal,
+              redirectChecker,
+              depth + 1,
+            )
+          } else {
+            return {
+              type: 'redirect' as const,
+              originalUrl: url,
+              redirectUrl,
+              statusCode: fetchResponse.status,
+            }
+          }
+        }
+        const arrayBuffer = await fetchResponse.arrayBuffer()
+        // Build an AxiosResponse-like shape so downstream code stays happy
+        return {
+          data: new Uint8Array(arrayBuffer),
+          status: fetchResponse.status,
+          statusText: fetchResponse.statusText,
+          headers: Object.fromEntries(fetchResponse.headers.entries()),
+          config: axiosConfig,
+          request: undefined,
+        } as unknown as AxiosResponse<ArrayBuffer>
+      } catch {
+        // Fall through to original error handling
+      }
+    }
+
     if (
       axios.isAxiosError(error) &&
       error.response &&
@@ -489,6 +545,58 @@ export async function getURLMarkdownContent(
   return entry
 }
 
+// Budget for the secondary-model summarization after fetch. If the small-
+// fast model is slow (e.g. a 200k-context third-party running a reasoning
+// pass over ~100KB of markdown), we'd rather fall back to raw truncated
+// markdown than hang the tool. Also keeps the worst-case WebFetch bounded
+// to FETCH_TIMEOUT_MS + SECONDARY_MODEL_TIMEOUT_MS regardless of provider.
+const SECONDARY_MODEL_TIMEOUT_MS = 45_000
+
+function raceWithTimeout<T>(
+  promise: Promise<T>,
+  timeoutMs: number,
+  signal: AbortSignal,
+): Promise<T> {
+  return new Promise<T>((resolve, reject) => {
+    const timer = setTimeout(() => {
+      const err = new Error(`Secondary-model summarization timed out after ${timeoutMs}ms`)
+      ;(err as NodeJS.ErrnoException).code = 'SECONDARY_MODEL_TIMEOUT'
+      reject(err)
+    }, timeoutMs)
+    const onAbort = () => {
+      clearTimeout(timer)
+      reject(new AbortError())
+    }
+    if (signal.aborted) {
+      clearTimeout(timer)
+      reject(new AbortError())
+      return
+    }
+    signal.addEventListener('abort', onAbort, { once: true })
+    promise.then(
+      value => {
+        clearTimeout(timer)
+        signal.removeEventListener('abort', onAbort)
+        resolve(value)
+      },
+      err => {
+        clearTimeout(timer)
+        signal.removeEventListener('abort', onAbort)
+        reject(err)
+      },
+    )
+  })
+}
+
+function buildFallbackMarkdownSummary(truncatedContent: string): string {
+  return [
+    '[Secondary-model summarization unavailable — returning raw fetched content.',
+    'This typically means the configured small-fast model took too long or errored.]',
+    '',
+    truncatedContent,
+  ].join('\n')
+}
+
 export async function applyPromptToMarkdown(
   prompt: string,
   markdownContent: string,
@@ -508,18 +616,35 @@ export async function applyPromptToMarkdown(
     prompt,
     isPreapprovedDomain,
   )
-  const assistantMessage = await queryHaiku({
+  let assistantMessage
-    systemPrompt: asSystemPrompt([]),
+  try {
-    userPrompt: modelPrompt,
+    assistantMessage = await raceWithTimeout(
-    signal,
+      queryHaiku({
-    options: {
+        systemPrompt: asSystemPrompt([]),
-      querySource: 'web_fetch_apply',
+        userPrompt: modelPrompt,
-      agents: [],
+        signal,
-      isNonInteractiveSession,
+        options: {
-      hasAppendSystemPrompt: false,
+          querySource: 'web_fetch_apply',
-      mcpTools: [],
+          agents: [],
-    },
+          isNonInteractiveSession,
-  })
+          hasAppendSystemPrompt: false,
+          mcpTools: [],
+        },
+      }),
+      SECONDARY_MODEL_TIMEOUT_MS,
+      signal,
+    )
+  } catch (err) {
+    // User interrupts and SIGINTs still propagate. Everything else (timeout,
+    // provider-side error, unsupported model on third-party endpoint) falls
+    // back to raw markdown so the user still gets usable content rather than
+    // a hang. Log so it's visible in debug traces.
+    if (err instanceof AbortError || (err as Error)?.name === 'AbortError') {
+      throw err
+    }
+    logError(err)
+    return buildFallbackMarkdownSummary(truncatedContent)
+  }
 
   // We need to bubble this up, so that the tool call throws, causing us to return
   // an is_error tool_use block to the server, and render a red dot in the UI.
@@ -534,5 +659,5 @@ export async function applyPromptToMarkdown(
       return contentBlock.text
     }
   }
-  return 'No response from model'
+  return buildFallbackMarkdownSummary(truncatedContent)
 }

src/tools/WebSearchTool/WebSearchTool.ts

@@ -203,6 +203,61 @@ function buildCodexWebSearchInstructions(): string {
   ].join(' ')
 }
 
+function pushCodexTextResult(
+  results: (SearchResult | string)[],
+  value: unknown,
+): void {
+  if (typeof value !== 'string') return
+  const trimmed = value.trim()
+  if (trimmed) {
+    results.push(trimmed)
+  }
+}
+
+function addCodexSource(
+  sourceMap: Map<string, { title: string; url: string }>,
+  source: unknown,
+): void {
+  if (typeof source?.url !== 'string' || !source.url) return
+  sourceMap.set(source.url, {
+    title:
+      typeof source.title === 'string' && source.title
+        ? source.title
+        : source.url,
+    url: source.url,
+  })
+}
+
+function getCodexSources(item: Record<string, any>): unknown[] {
+  if (Array.isArray(item.action?.sources)) {
+    return item.action.sources
+  }
+  if (Array.isArray(item.sources)) {
+    return item.sources
+  }
+  if (Array.isArray(item.result?.sources)) {
+    return item.result.sources
+  }
+  return []
+}
+
+function extractCodexWebSearchFailure(item: Record<string, any>): string | undefined {
+  // Codex web_search_call items can carry a status field. When the tool
+  // call fails (rate limit, upstream error, model-side guardrail), the
+  // parser should surface a meaningful error rather than the generic
+  // "No results found." fallback. Shape observed across recent payloads:
+  //   { type: 'web_search_call', status: 'failed', error: { message?: string } }
+  //   { type: 'web_search_call', status: 'failed', action: { error?: { message?: string } } }
+  if (item?.status !== 'failed') return undefined
+  const reason =
+    (typeof item.error?.message === 'string' && item.error.message) ||
+    (typeof item.action?.error?.message === 'string' &&
+      item.action.error.message) ||
+    (typeof item.error === 'string' && item.error) ||
+    undefined
+  return reason ? `Web search failed: ${reason}` : 'Web search failed.'
+}
+
 function makeOutputFromCodexWebSearchResponse(
   response: Record<string, unknown>,
   query: string,
@@ -214,18 +269,12 @@ function makeOutputFromCodexWebSearchResponse(
 
   for (const item of output) {
     if (item?.type === 'web_search_call') {
-      const sources = Array.isArray(item.action?.sources)
+      const failure = extractCodexWebSearchFailure(item)
-        ? item.action.sources
+      if (failure) {
-        : []
+        results.push(failure)
-      for (const source of sources) {
+      }
-        if (typeof source?.url !== 'string' || !source.url) continue
+      for (const source of getCodexSources(item)) {
-        sourceMap.set(source.url, {
+        addCodexSource(sourceMap, source)
-          title:
-            typeof source.title === 'string' && source.title
-              ? source.title
-              : source.url,
-          url: source.url,
-        })
       }
       continue
     }
@@ -235,11 +284,12 @@ function makeOutputFromCodexWebSearchResponse(
     }
 
     for (const part of item.content) {
-      if (part?.type === 'output_text' && typeof part.text === 'string') {
+      if (part?.type === 'output_text' || part?.type === 'text') {
-        const trimmed = part.text.trim()
+        pushCodexTextResult(results, part.text)
-        if (trimmed) {
+      }
-          results.push(trimmed)
+
-        }
+      for (const source of getCodexSources(part)) {
+        addCodexSource(sourceMap, source)
       }
 
       const annotations = Array.isArray(part?.annotations)
@@ -247,23 +297,13 @@ function makeOutputFromCodexWebSearchResponse(
         : []
       for (const annotation of annotations) {
         if (annotation?.type !== 'url_citation') continue
-        if (typeof annotation.url !== 'string' || !annotation.url) continue
+        addCodexSource(sourceMap, annotation)
-        sourceMap.set(annotation.url, {
-          title:
-            typeof annotation.title === 'string' && annotation.title
-              ? annotation.title
-              : annotation.url,
-          url: annotation.url,
-        })
       }
     }
   }
 
-  if (results.length === 0 && typeof response.output_text === 'string') {
+  if (results.length === 0) {
-    const trimmed = response.output_text.trim()
+    pushCodexTextResult(results, response.output_text)
-    if (trimmed) {
-      results.push(trimmed)
-    }
   }
 
   if (sourceMap.size > 0) {
@@ -273,6 +313,10 @@ function makeOutputFromCodexWebSearchResponse(
     })
   }
 
+  if (results.length === 0) {
+    results.push('No results found.')
+  }
+
   return {
     query,
     results,
@@ -280,6 +324,10 @@ function makeOutputFromCodexWebSearchResponse(
   }
 }
 
+export const __test = {
+  makeOutputFromCodexWebSearchResponse,
+}
+
 async function runCodexWebSearch(
   input: Input,
   signal: AbortSignal,
@@ -457,6 +505,19 @@ function shouldUseAdapterProvider(): boolean {
   return getAvailableProviders().length > 0
 }
 
+/**
+ * Returns true when the current provider has a working native or Codex
+ * web-search fallback after an adapter failure. OpenAI shim providers
+ * (moonshot, minimax, nvidia-nim, openai, github, etc.) do NOT support
+ * Anthropic's web_search_20250305 tool, so falling through to the native
+ * path silently produces "Did 0 searches".
+ */
+function hasNativeSearchFallback(): boolean {
+  if (isCodexResponsesWebSearchEnabled()) return true
+  const provider = getAPIProvider()
+  return provider === 'firstParty' || provider === 'vertex' || provider === 'foundry'
+}
+
 // ---------------------------------------------------------------------------
 // Tool export
 // ---------------------------------------------------------------------------
@@ -609,6 +670,17 @@ export const WebSearchTool = buildTool({
         // Auto mode: only fall through on transient errors (network, timeout, 5xx).
         // Config / guardrail errors (SSRF, HTTPS, bad URL, etc.) must surface.
         if (!isTransientError(err)) throw err
+        // No viable fallback for this provider — surface the adapter error
+        // instead of falling through to a broken native path.
+        if (!hasNativeSearchFallback()) {
+          const provider = getAPIProvider()
+          const errMsg = err instanceof Error ? err.message : String(err)
+          throw new Error(
+            `Web search is unavailable for provider "${provider}". ` +
+              `The search adapter failed (${errMsg}). ` +
+              `Try switching to a provider with built-in web search (e.g. Anthropic, Codex) or try again later.`,
+          )
+        }
         console.error(
           `[web-search] Adapter failed, falling through to native: ${err}`,
         )

src/tools/WebSearchTool/providers/duckduckgo.ts

@@ -1,6 +1,44 @@
 import type { SearchInput, SearchProvider } from './types.js'
 import { applyDomainFilters, type ProviderOutput } from './types.js'
 
+// DuckDuckGo's HTML scraper aggressively blocks datacenter / repeat IPs with
+// an "anomaly in the request" response. When that happens we surface an
+// actionable error instead of the opaque scraper message so users know how
+// to configure a working backend.
+const DDG_ANOMALY_HINT =
+  'DuckDuckGo scraping is rate-limited from this network. ' +
+  'Configure a search backend with one of: ' +
+  'FIRECRAWL_API_KEY, TAVILY_API_KEY, EXA_API_KEY, YOU_API_KEY, ' +
+  'JINA_API_KEY, BING_API_KEY, MOJEEK_API_KEY, LINKUP_API_KEY — ' +
+  'or use an Anthropic / Vertex / Foundry provider for native web search.'
+
+const MAX_RETRIES = 3
+const INITIAL_BACKOFF_MS = 1000
+
+function isAnomalyError(message: string): boolean {
+  return /anomaly in the request|likely making requests too quickly/i.test(
+    message,
+  )
+}
+
+function isRetryableDDGError(err: unknown): boolean {
+  if (!(err instanceof Error)) return false
+  const msg = err.message.toLowerCase()
+  return (
+    msg.includes('anomaly') ||
+    msg.includes('too quickly') ||
+    msg.includes('rate limit') ||
+    msg.includes('timeout') ||
+    msg.includes('econnreset') ||
+    msg.includes('etimedout') ||
+    msg.includes('econnaborted')
+  )
+}
+
+function sleep(ms: number): Promise<void> {
+  return new Promise(r => setTimeout(r, ms))
+}
+
 export const duckduckgoProvider: SearchProvider = {
   name: 'duckduckgo',
 
@@ -19,22 +57,44 @@ export const duckduckgoProvider: SearchProvider = {
       throw new Error('duck-duck-scrape package not installed. Run: npm install duck-duck-scrape')
     }
     if (signal?.aborted) throw new DOMException('Aborted', 'AbortError')
-    // TODO: duck-duck-scrape doesn't accept AbortSignal — can't cancel in-flight searches
-    const response = await search(input.query, { safeSearch: SafeSearchType.STRICT })
 
-    const hits = applyDomainFilters(
+    let lastErr: unknown
-      response.results.map(r => ({
+    for (let attempt = 0; attempt < MAX_RETRIES; attempt++) {
-        title: r.title || r.url,
+      if (signal?.aborted) throw new DOMException('Aborted', 'AbortError')
-        url: r.url,
+      try {
-        description: r.description ?? undefined,
+        // TODO: duck-duck-scrape doesn't accept AbortSignal — can't cancel in-flight searches
-      })),
+        const response = await search(input.query, { safeSearch: SafeSearchType.STRICT })
-      input,
-    )
 
-    return {
+        const hits = applyDomainFilters(
-      hits,
+          response.results.map(r => ({
-      providerName: 'duckduckgo',
+            title: r.title || r.url,
-      durationSeconds: (performance.now() - start) / 1000,
+            url: r.url,
+            description: r.description ?? undefined,
+          })),
+          input,
+        )
+
+        return {
+          hits,
+          providerName: 'duckduckgo',
+          durationSeconds: (performance.now() - start) / 1000,
+        }
+      } catch (err) {
+        lastErr = err
+        const msg = err instanceof Error ? err.message : String(err)
+        if (isAnomalyError(msg)) {
+          throw new Error(DDG_ANOMALY_HINT)
+        }
+        if (!isRetryableDDGError(err) || attempt === MAX_RETRIES - 1) {
+          throw err
+        }
+        // Exponential backoff with jitter: 1s, 2s, 4s +/- 20%
+        const baseDelay = INITIAL_BACKOFF_MS * Math.pow(2, attempt)
+        const jitter = baseDelay * 0.2 * (Math.random() * 2 - 1)
+        await sleep(baseDelay + jitter)
+      }
     }
+
+    throw lastErr
   },
 }

src/utils/context.ts

@@ -12,7 +12,12 @@ export const MODEL_CONTEXT_WINDOW_DEFAULT = 200_000
 // Fallback context window for unknown 3P models. Must be large enough that
 // the effective context (this minus output token reservation) stays positive,
 // otherwise auto-compact fires on every message (issue #635).
-export const OPENAI_FALLBACK_CONTEXT_WINDOW = 128_000
+// Override via CLAUDE_CODE_OPENAI_FALLBACK_CONTEXT_WINDOW env var to avoid
+// hardcoding when deploying models not yet in openaiContextWindows.ts.
+export const OPENAI_FALLBACK_CONTEXT_WINDOW = (() => {
+  const v = parseInt(process.env.CLAUDE_CODE_OPENAI_FALLBACK_CONTEXT_WINDOW ?? '', 10)
+  return !isNaN(v) && v > 0 ? v : 128_000
+})()
 
 // Maximum output tokens for compact operations
 export const COMPACT_MAX_OUTPUT_TOKENS = 20_000

src/utils/hookChains.integration.test.ts

@@ -0,0 +1,357 @@
+import { afterEach, beforeEach, describe, expect, mock, test } from 'bun:test'
+import { mkdtemp, rm, writeFile } from 'node:fs/promises'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+
+type HookChainsModule = typeof import('./hookChains.js')
+
+type ImportHarnessOptions = {
+  allowRemoteSessions?: boolean
+  teamFile?:
+    | {
+        name: string
+        members: Array<{ name: string }>
+      }
+    | null
+  teamName?: string
+  senderName?: string
+  replBridgeHandle?: unknown
+}
+
+const tempDirs: string[] = []
+const originalHookChainsEnabled = process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS
+
+async function createConfigFile(config: unknown): Promise<string> {
+  const dir = await mkdtemp(join(tmpdir(), 'openclaude-hook-chains-int-'))
+  tempDirs.push(dir)
+  const filePath = join(dir, 'hook-chains.json')
+  await writeFile(filePath, JSON.stringify(config, null, 2), 'utf-8')
+  return filePath
+}
+
+async function importHookChainsHarness(
+  options: ImportHarnessOptions = {},
+): Promise<{
+  mod: HookChainsModule
+  writeToMailboxSpy: ReturnType<typeof mock>
+  agentToolCallSpy: ReturnType<typeof mock>
+}> {
+  mock.restore()
+
+  const allowRemoteSessions = options.allowRemoteSessions ?? true
+  const teamName = options.teamName ?? 'mesh-team'
+  const senderName = options.senderName ?? 'mesh-lead'
+  const replBridgeHandle = options.replBridgeHandle ?? null
+
+  const writeToMailboxSpy = mock(async () => {})
+  const agentToolCallSpy = mock(async () => ({
+    data: {
+      status: 'async_launched',
+      agentId: 'agent-fallback-1',
+    },
+  }))
+
+  mock.module('../services/analytics/index.js', () => ({
+    logEvent: () => {},
+  }))
+
+  mock.module('./telemetry/events.js', () => ({
+    logOTelEvent: async () => {},
+  }))
+
+  mock.module('../services/policyLimits/index.js', () => ({
+    isPolicyAllowed: () => allowRemoteSessions,
+  }))
+
+  mock.module('./swarm/teamHelpers.js', () => ({
+    readTeamFileAsync: async () => options.teamFile ?? null,
+  }))
+
+  mock.module('./teammateMailbox.js', () => ({
+    writeToMailbox: writeToMailboxSpy,
+  }))
+
+  mock.module('./teammate.js', () => ({
+    getAgentName: () => senderName,
+    getTeamName: () => teamName,
+    getTeammateColor: () => 'blue',
+    // Keep parity with the real module's surface so later tests that
+    // run after this file (mock.module is process-global and mock.restore
+    // does not undo module mocks in Bun) do not see undefined members.
+    isTeammate: () => false,
+    isPlanModeRequired: () => false,
+    getAgentId: () => undefined,
+    getParentSessionId: () => undefined,
+  }))
+
+  mock.module('../bridge/replBridgeHandle.js', () => ({
+    getReplBridgeHandle: () => replBridgeHandle,
+  }))
+
+  // Integration mock target requested in the task: fallback action can route
+  // through this mocked tool launcher from runtime callback wiring.
+  mock.module('../tools/AgentTool/AgentTool.js', () => ({
+    AgentTool: {
+      call: agentToolCallSpy,
+    },
+  }))
+
+  const mod = await import(`./hookChains.js?integration=${Date.now()}-${Math.random()}`)
+  return { mod, writeToMailboxSpy, agentToolCallSpy }
+}
+
+beforeEach(() => {
+  process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS = '1'
+})
+
+afterEach(async () => {
+  mock.restore()
+
+  if (originalHookChainsEnabled === undefined) {
+    delete process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS
+  } else {
+    process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS = originalHookChainsEnabled
+  }
+
+  await Promise.all(
+    tempDirs.splice(0).map(dir => rm(dir, { recursive: true, force: true })),
+  )
+})
+
+describe('hookChains integration dispatch', () => {
+  test('end-to-end rule evaluation + action dispatch on TaskCompleted failure', async () => {
+    const { mod } = await importHookChainsHarness({
+      teamName: 'mesh-team',
+      senderName: 'mesh-lead',
+      teamFile: {
+        name: 'mesh-team',
+        members: [{ name: 'mesh-lead' }, { name: 'worker-a' }, { name: 'worker-b' }],
+      },
+    })
+
+    const configPath = await createConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'task-failure-recovery',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [
+            { type: 'spawn_fallback_agent' },
+            { type: 'notify_team' },
+          ],
+        },
+      ],
+    })
+
+    const spawnSpy = mock(async () => ({ launched: true, agentId: 'agent-e2e-1' }))
+    const notifySpy = mock(async () => ({ sent: true, recipientCount: 2 }))
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: {
+          task_id: 'task-001',
+          task_subject: 'Patch flaky build',
+          error: 'CI timeout',
+        },
+      },
+      runtime: {
+        onSpawnFallbackAgent: spawnSpy,
+        onNotifyTeam: notifySpy,
+      },
+    })
+
+    expect(result.enabled).toBe(true)
+    expect(result.matchedRuleIds).toEqual(['task-failure-recovery'])
+    expect(result.actionResults).toHaveLength(2)
+    expect(result.actionResults[0]?.status).toBe('executed')
+    expect(result.actionResults[1]?.status).toBe('executed')
+    expect(spawnSpy).toHaveBeenCalledTimes(1)
+    expect(notifySpy).toHaveBeenCalledTimes(1)
+  })
+
+  test('fallback spawn injects failure context into generated prompt', async () => {
+    const { mod, agentToolCallSpy } = await importHookChainsHarness()
+
+    const configPath = await createConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'fallback-context',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [
+            {
+              type: 'spawn_fallback_agent',
+              description: 'Fallback for failed task',
+            },
+          ],
+        },
+      ],
+    })
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: {
+          task_id: 'task-ctx-1',
+          task_subject: 'Repair migration guard',
+          task_description: 'Fix regression in check ordering',
+          error: 'Task failed after retry budget exhausted',
+        },
+      },
+      runtime: {
+        onSpawnFallbackAgent: async request => {
+          const { AgentTool } = await import('../tools/AgentTool/AgentTool.js')
+          await (AgentTool.call as unknown as (...args: unknown[]) => Promise<unknown>)({
+            prompt: request.prompt,
+            description: request.description,
+            run_in_background: request.runInBackground,
+            subagent_type: request.agentType,
+            model: request.model,
+          })
+          return { launched: true, agentId: 'agent-fallback-ctx' }
+        },
+      },
+    })
+
+    expect(result.actionResults[0]?.status).toBe('executed')
+    expect(agentToolCallSpy).toHaveBeenCalledTimes(1)
+
+    const callInput = agentToolCallSpy.mock.calls[0]?.[0] as {
+      prompt: string
+      description: string
+      run_in_background: boolean
+    }
+
+    expect(callInput.description).toBe('Fallback for failed task')
+    expect(callInput.run_in_background).toBe(true)
+    expect(callInput.prompt).toContain('Event: TaskCompleted')
+    expect(callInput.prompt).toContain('Outcome: failed')
+    expect(callInput.prompt).toContain('Task subject: Repair migration guard')
+    expect(callInput.prompt).toContain('Failure details: Task failed after retry budget exhausted')
+  })
+
+  test('notify_team dispatches mailbox writes when team exists and skips when absent', async () => {
+    const withTeam = await importHookChainsHarness({
+      teamName: 'mesh-a',
+      senderName: 'lead-a',
+      teamFile: {
+        name: 'mesh-a',
+        members: [{ name: 'lead-a' }, { name: 'worker-1' }, { name: 'worker-2' }],
+      },
+    })
+
+    const configPathWithTeam = await createConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'notify-existing-team',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'notify_team' }],
+        },
+      ],
+    })
+
+    const withTeamResult = await withTeam.mod.dispatchHookChainsForEvent({
+      configPathOverride: configPathWithTeam,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-team-ok', error: 'boom' },
+      },
+    })
+
+    expect(withTeamResult.actionResults[0]?.status).toBe('executed')
+    expect(withTeam.writeToMailboxSpy).toHaveBeenCalledTimes(2)
+
+    const recipients = withTeam.writeToMailboxSpy.mock.calls.map(
+      call => call[0] as string,
+    )
+    expect(recipients.sort()).toEqual(['worker-1', 'worker-2'])
+
+    const withoutTeam = await importHookChainsHarness({
+      teamName: 'mesh-missing',
+      senderName: 'lead-missing',
+      teamFile: null,
+    })
+
+    const configPathWithoutTeam = await createConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'notify-missing-team',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'notify_team' }],
+        },
+      ],
+    })
+
+    const withoutTeamResult = await withoutTeam.mod.dispatchHookChainsForEvent({
+      configPathOverride: configPathWithoutTeam,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-team-missing', error: 'boom' },
+      },
+    })
+
+    expect(withoutTeamResult.actionResults[0]?.status).toBe('skipped')
+    expect(withoutTeamResult.actionResults[0]?.reason).toContain('Team file not found')
+    expect(withoutTeam.writeToMailboxSpy).not.toHaveBeenCalled()
+  })
+
+  test('warm_remote_capacity is a safe no-op when bridge is inactive', async () => {
+    const { mod } = await importHookChainsHarness({
+      allowRemoteSessions: true,
+      replBridgeHandle: null,
+    })
+
+    const configPath = await createConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'bridge-warmup-noop',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'warm_remote_capacity' }],
+        },
+      ],
+    })
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-warm-1' },
+      },
+    })
+
+    expect(result.actionResults).toHaveLength(1)
+    expect(result.actionResults[0]?.status).toBe('skipped')
+    expect(result.actionResults[0]?.reason).toContain('Bridge is not active')
+  })
+})

src/utils/hookChains.test.ts

@@ -0,0 +1,476 @@
+import { afterEach, beforeEach, describe, expect, mock, test } from 'bun:test'
+import { mkdtemp, rm, writeFile } from 'node:fs/promises'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+
+type HookChainsModule = typeof import('./hookChains.js')
+
+const tempDirs: string[] = []
+const originalHookChainsEnabled = process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS
+
+async function makeConfigFile(config: unknown): Promise<string> {
+  const dir = await mkdtemp(join(tmpdir(), 'openclaude-hook-chains-'))
+  tempDirs.push(dir)
+  const filePath = join(dir, 'hook-chains.json')
+  await writeFile(filePath, JSON.stringify(config, null, 2), 'utf-8')
+  return filePath
+}
+
+async function importHookChainsModule(options?: {
+  allowRemoteSessions?: boolean
+}): Promise<HookChainsModule> {
+  mock.restore()
+
+  const allowRemoteSessions = options?.allowRemoteSessions ?? true
+
+  mock.module('../services/analytics/index.js', () => ({
+    logEvent: () => {},
+  }))
+
+  mock.module('./telemetry/events.js', () => ({
+    logOTelEvent: async () => {},
+  }))
+
+  mock.module('../services/policyLimits/index.js', () => ({
+    isPolicyAllowed: () => allowRemoteSessions,
+  }))
+
+  return import(`./hookChains.js?test=${Date.now()}-${Math.random()}`)
+}
+
+beforeEach(() => {
+  process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS = '1'
+})
+
+afterEach(async () => {
+  mock.restore()
+
+  if (originalHookChainsEnabled === undefined) {
+    delete process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS
+  } else {
+    process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS = originalHookChainsEnabled
+  }
+
+  await Promise.all(
+    tempDirs.splice(0).map(dir => rm(dir, { recursive: true, force: true })),
+  )
+})
+
+describe('hookChains schema validation', () => {
+  test('returns disabled config when env gate is unset', async () => {
+    delete process.env.CLAUDE_CODE_ENABLE_HOOK_CHAINS
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      rules: [
+        {
+          id: 'env-gated-rule',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const loaded = mod.loadHookChainsConfig({ pathOverride: configPath })
+    expect(loaded.exists).toBe(false)
+    expect(loaded.config.enabled).toBe(false)
+    expect(loaded.config.rules).toHaveLength(0)
+  })
+
+  test('loads valid config and memoizes by mtime/size', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 5000,
+      defaultDedupWindowMs: 5000,
+      rules: [
+        {
+          id: 'task-failure-fallback',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [
+            {
+              type: 'spawn_fallback_agent',
+              description: 'Fallback recovery agent',
+            },
+          ],
+        },
+      ],
+    })
+
+    const first = mod.loadHookChainsConfig({ pathOverride: configPath })
+    expect(first.exists).toBe(true)
+    expect(first.error).toBeUndefined()
+    expect(first.fromCache).toBe(false)
+    expect(first.config.enabled).toBe(true)
+    expect(first.config.rules).toHaveLength(1)
+    expect(first.config.rules[0]?.id).toBe('task-failure-fallback')
+
+    const second = mod.loadHookChainsConfig({ pathOverride: configPath })
+    expect(second.exists).toBe(true)
+    expect(second.error).toBeUndefined()
+    expect(second.fromCache).toBe(true)
+    expect(second.config.rules).toHaveLength(1)
+  })
+
+  test('accepts wrapped { hookChains: ... } config shape', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      hookChains: {
+        version: 1,
+        enabled: true,
+        rules: [
+          {
+            id: 'wrapped-shape',
+            trigger: { event: 'PostToolUseFailure', outcomes: ['failed'] },
+            actions: [{ type: 'notify_team' }],
+          },
+        ],
+      },
+    })
+
+    const loaded = mod.loadHookChainsConfig({ pathOverride: configPath })
+    expect(loaded.error).toBeUndefined()
+    expect(loaded.config.enabled).toBe(true)
+    expect(loaded.config.rules[0]?.id).toBe('wrapped-shape')
+  })
+
+  test('returns disabled config for invalid schema', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      rules: [
+        {
+          id: 'invalid-rule',
+          trigger: {
+            event: 'TaskCompleted',
+            outcome: 'failed',
+            outcomes: ['failed'],
+          },
+          actions: [{ type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const loaded = mod.loadHookChainsConfig({ pathOverride: configPath })
+    expect(loaded.exists).toBe(true)
+    expect(loaded.error).toBeDefined()
+    expect(loaded.config.enabled).toBe(false)
+    expect(loaded.config.rules).toHaveLength(0)
+  })
+})
+
+describe('evaluateHookChainRules', () => {
+  test('matches by event + outcome + condition', async () => {
+    const mod = await importHookChainsModule()
+
+    const rules = [
+      {
+        id: 'post-tool-failure-rule',
+        trigger: { event: 'PostToolUseFailure', outcome: 'failed' },
+        condition: {
+          toolNames: ['Edit'],
+          errorIncludes: ['permission'],
+          eventFieldEquals: { 'meta.source': 'scheduler' },
+        },
+        actions: [{ type: 'spawn_fallback_agent' }],
+      },
+    ]
+
+    const matches = mod.evaluateHookChainRules(rules as never, {
+      eventName: 'PostToolUseFailure',
+      outcome: 'failed',
+      payload: {
+        tool_name: 'Edit',
+        error: 'Permission denied by policy',
+        meta: { source: 'scheduler' },
+      },
+    })
+
+    expect(matches).toHaveLength(1)
+    expect(matches[0]?.rule.id).toBe('post-tool-failure-rule')
+  })
+
+  test('does not match when event/condition fail', async () => {
+    const mod = await importHookChainsModule()
+
+    const rules = [
+      {
+        id: 'rule-no-match',
+        trigger: { event: 'PostToolUseFailure', outcomes: ['failed'] },
+        condition: { toolNames: ['Write'] },
+        actions: [{ type: 'spawn_fallback_agent' }],
+      },
+    ]
+
+    const wrongEvent = mod.evaluateHookChainRules(rules as never, {
+      eventName: 'TaskCompleted',
+      outcome: 'failed',
+      payload: { tool_name: 'Write' },
+    })
+    expect(wrongEvent).toHaveLength(0)
+
+    const wrongCondition = mod.evaluateHookChainRules(rules as never, {
+      eventName: 'PostToolUseFailure',
+      outcome: 'failed',
+      payload: { tool_name: 'Edit' },
+    })
+    expect(wrongCondition).toHaveLength(0)
+  })
+})
+
+describe('dispatchHookChainsForEvent guard logic', () => {
+  test('dedup skips duplicate event/action within dedup window', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 4,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 60_000,
+      rules: [
+        {
+          id: 'dedup-rule',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          cooldownMs: 0,
+          dedupWindowMs: 60_000,
+          actions: [{ id: 'spawn-1', type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const spawn = mock(async () => ({ launched: true, agentId: 'agent-1' }))
+
+    const first = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-123', error: 'boom' },
+      },
+      runtime: { onSpawnFallbackAgent: spawn },
+    })
+
+    const second = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-123', error: 'boom' },
+      },
+      runtime: { onSpawnFallbackAgent: spawn },
+    })
+
+    expect(first.actionResults[0]?.status).toBe('executed')
+    expect(second.actionResults[0]?.status).toBe('skipped')
+    expect(second.actionResults[0]?.reason).toContain('dedup')
+    expect(spawn).toHaveBeenCalledTimes(1)
+  })
+
+  test('cooldown skips second dispatch when rule cooldown is active', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 4,
+      defaultCooldownMs: 60_000,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'cooldown-rule',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          cooldownMs: 60_000,
+          dedupWindowMs: 0,
+          actions: [{ type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const spawn = mock(async () => ({ launched: true, agentId: 'agent-2' }))
+
+    const first = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-456' },
+      },
+      runtime: { onSpawnFallbackAgent: spawn },
+    })
+
+    const second = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-789' },
+      },
+      runtime: { onSpawnFallbackAgent: spawn },
+    })
+
+    expect(first.actionResults[0]?.status).toBe('executed')
+    expect(second.actionResults[0]?.status).toBe('skipped')
+    expect(second.actionResults[0]?.reason).toContain('cooldown')
+    expect(spawn).toHaveBeenCalledTimes(1)
+  })
+
+  test('depth limit blocks dispatch when chain depth reaches max', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 1,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'depth-rule',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const spawn = mock(async () => ({ launched: true, agentId: 'agent-3' }))
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-depth' },
+      },
+      runtime: {
+        chainDepth: 1,
+        onSpawnFallbackAgent: spawn,
+      },
+    })
+
+    expect(result.enabled).toBe(true)
+    expect(result.matchedRuleIds).toHaveLength(0)
+    expect(result.actionResults).toHaveLength(0)
+    expect(spawn).not.toHaveBeenCalled()
+  })
+})
+
+describe('action dispatch skip scenarios', () => {
+  test('fails spawn_fallback_agent when launcher callback is missing', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'missing-launcher',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'spawn_fallback_agent' }],
+        },
+      ],
+    })
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-missing-launcher' },
+      },
+      runtime: {},
+    })
+
+    expect(result.actionResults[0]?.status).toBe('failed')
+    expect(result.actionResults[0]?.reason).toContain('launcher')
+  })
+
+  test('skips disabled action and does not execute callback', async () => {
+    const mod = await importHookChainsModule()
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'disabled-action-rule',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [
+            {
+              type: 'spawn_fallback_agent',
+              enabled: false,
+            },
+          ],
+        },
+      ],
+    })
+
+    const spawn = mock(async () => ({ launched: true, agentId: 'agent-4' }))
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-disabled' },
+      },
+      runtime: { onSpawnFallbackAgent: spawn },
+    })
+
+    expect(result.actionResults[0]?.status).toBe('skipped')
+    expect(result.actionResults[0]?.reason).toContain('disabled')
+    expect(spawn).not.toHaveBeenCalled()
+  })
+
+  test('skips warm_remote_capacity when policy denies remote sessions', async () => {
+    const mod = await importHookChainsModule({ allowRemoteSessions: false })
+
+    const configPath = await makeConfigFile({
+      version: 1,
+      enabled: true,
+      maxChainDepth: 3,
+      defaultCooldownMs: 0,
+      defaultDedupWindowMs: 0,
+      rules: [
+        {
+          id: 'policy-denied-remote-warm',
+          trigger: { event: 'TaskCompleted', outcome: 'failed' },
+          actions: [{ type: 'warm_remote_capacity' }],
+        },
+      ],
+    })
+
+    const warm = mock(async () => ({
+      warmed: true,
+      environmentId: 'env-123',
+    }))
+
+    const result = await mod.dispatchHookChainsForEvent({
+      configPathOverride: configPath,
+      event: {
+        eventName: 'TaskCompleted',
+        outcome: 'failed',
+        payload: { task_id: 'task-policy-denied' },
+      },
+      runtime: { onWarmRemoteCapacity: warm },
+    })
+
+    expect(result.actionResults[0]?.status).toBe('skipped')
+    expect(result.actionResults[0]?.reason).toContain('policy')
+    expect(warm).not.toHaveBeenCalled()
+  })
+})

src/utils/hooks.ts

@@ -10,6 +10,7 @@ import { wrapSpawn } from './ShellCommand.js'
 import { TaskOutput } from './task/TaskOutput.js'
 import { getCwd } from './cwd.js'
 import { randomUUID } from 'crypto'
+import { feature } from 'bun:bundle'
 import { formatShellPrefixCommand } from './bash/shellPrefix.js'
 import {
   getHookEnvFilePath,
@@ -134,6 +135,7 @@ import { registerPendingAsyncHook } from './hooks/AsyncHookRegistry.js'
 import { enqueuePendingNotification } from './messageQueueManager.js'
 import {
   extractTextContent,
+  createAssistantMessage,
   getLastAssistantMessage,
   wrapInSystemReminder,
 } from './messages.js'
@@ -145,6 +147,7 @@ import {
 import { createAttachmentMessage } from './attachments.js'
 import { all } from './generators.js'
 import { findToolByName, type Tools, type ToolUseContext } from '../Tool.js'
+import type { CanUseToolFn } from '../hooks/useCanUseTool.js'
 import { execPromptHook } from './hooks/execPromptHook.js'
 import type { Message, AssistantMessage } from '../types/message.js'
 import { execAgentHook } from './hooks/execAgentHook.js'
@@ -162,9 +165,147 @@ import type { AppState } from '../state/AppState.js'
 import { jsonStringify, jsonParse } from './slowOperations.js'
 import { isEnvTruthy } from './envUtils.js'
 import { errorMessage, getErrnoCode } from './errors.js'
+import { getAgentName, getTeamName, getTeammateColor } from './teammate.js'
+import type {
+  HookChainOutcome,
+  HookChainRuntimeContext,
+  SpawnFallbackAgentRequest,
+  SpawnFallbackAgentResponse,
+} from './hookChains.js'
 
 const TOOL_HOOK_EXECUTION_TIMEOUT_MS = 10 * 60 * 1000
 
+function normalizeFallbackAgentModel(
+  model: string | undefined,
+): 'sonnet' | 'opus' | 'haiku' | undefined {
+  if (model === 'sonnet' || model === 'opus' || model === 'haiku') {
+    return model
+  }
+  return undefined
+}
+
+async function launchFallbackAgentFromHookChains(
+  request: SpawnFallbackAgentRequest,
+  toolUseContext: ToolUseContext,
+  canUseTool: CanUseToolFn,
+): Promise<SpawnFallbackAgentResponse> {
+  try {
+    const { AgentTool } = await import('../tools/AgentTool/AgentTool.js')
+    const normalizedModel = normalizeFallbackAgentModel(request.model)
+    const result = await AgentTool.call(
+      {
+        prompt: request.prompt,
+        description: request.description,
+        run_in_background: true,
+        ...(request.agentType ? { subagent_type: request.agentType } : {}),
+        ...(normalizedModel ? { model: normalizedModel } : {}),
+      },
+      toolUseContext,
+      canUseTool,
+      createAssistantMessage({ content: [] }),
+    )
+
+    const data = result.data as
+      | {
+          status?: string
+          agentId?: string
+          agent_id?: string
+        }
+      | undefined
+    const status = data?.status
+
+    if (
+      status === 'async_launched' ||
+      status === 'completed' ||
+      status === 'remote_launched' ||
+      status === 'teammate_spawned'
+    ) {
+      return {
+        launched: true,
+        agentId: data?.agentId ?? data?.agent_id,
+      }
+    }
+
+    return {
+      launched: true,
+      reason:
+        status !== undefined
+          ? `Fallback launched with status ${status}`
+          : undefined,
+    }
+  } catch (error) {
+    return {
+      launched: false,
+      reason: `Fallback launch failed: ${errorMessage(error)}`,
+    }
+  }
+}
+
+async function dispatchHookChainFromHookRuntime(args: {
+  eventName: 'PostToolUseFailure' | 'TaskCompleted'
+  outcome: HookChainOutcome
+  payload: Record<string, unknown>
+  signal?: AbortSignal
+  toolUseContext?: ToolUseContext
+}): Promise<void> {
+  try {
+    if (!feature('HOOK_CHAINS')) {
+      return
+    }
+
+    const { dispatchHookChainsForEvent } = await import('./hookChains.js')
+
+    const runtime: HookChainRuntimeContext = {
+      signal: args.signal,
+      senderName: getAgentName() ?? undefined,
+      senderColor: getTeammateColor() ?? undefined,
+      teamName: getTeamName() ?? undefined,
+    }
+
+    const chainDepth = args.toolUseContext?.queryTracking?.depth
+    if (typeof chainDepth === 'number' && Number.isFinite(chainDepth)) {
+      runtime.chainDepth = chainDepth
+    }
+
+    const hookChainsCanUseTool = (
+      args.toolUseContext as
+        | (ToolUseContext & { hookChainsCanUseTool?: CanUseToolFn })
+        | undefined
+    )?.hookChainsCanUseTool
+
+    if (args.toolUseContext) {
+      runtime.onSpawnFallbackAgent = request => {
+        if (!hookChainsCanUseTool) {
+          return Promise.resolve({
+            launched: false,
+            reason:
+              'Fallback action requires canUseTool in this hook runtime context',
+          })
+        }
+
+        return launchFallbackAgentFromHookChains(
+          request,
+          args.toolUseContext!,
+          hookChainsCanUseTool,
+        )
+      }
+    }
+
+    await dispatchHookChainsForEvent({
+      event: {
+        eventName: args.eventName,
+        outcome: args.outcome,
+        payload: args.payload,
+      },
+      runtime,
+    })
+  } catch (error) {
+    logForDebugging(
+      `[hook-chains] Dispatch failed for ${args.eventName}: ${errorMessage(error)}`,
+    )
+  }
+}
+
 /**
  * SessionEnd hooks run during shutdown/clear and need a much tighter bound
  * than TOOL_HOOK_EXECUTION_TIMEOUT_MS. This value is used by callers as both
@@ -3502,9 +3643,11 @@ export async function* executePostToolUseFailureHooks<ToolInput>(
 ): AsyncGenerator<AggregatedHookResult> {
   const appState = toolUseContext.getAppState()
   const sessionId = toolUseContext.agentId ?? getSessionId()
-  if (!hasHookForEvent('PostToolUseFailure', appState, sessionId)) {
+  const hasPostToolFailureHooks = hasHookForEvent(
-    return
+    'PostToolUseFailure',
-  }
+    appState,
+    sessionId,
+  )
 
   const hookInput: PostToolUseFailureHookInput = {
     ...createBaseHookInput(permissionMode, undefined, toolUseContext),
@@ -3516,12 +3659,33 @@ export async function* executePostToolUseFailureHooks<ToolInput>(
     is_interrupt: isInterrupt,
   }
 
-  yield* executeHooks({
+  let blockingHookCount = 0
-    hookInput,
+
-    toolUseID,
+  if (hasPostToolFailureHooks) {
-    matchQuery: toolName,
+    for await (const result of executeHooks({
+      hookInput,
+      toolUseID,
+      matchQuery: toolName,
+      signal,
+      timeoutMs,
+      toolUseContext,
+    })) {
+      if (result.blockingError) {
+        blockingHookCount++
+      }
+      yield result
+    }
+  }
+
+  await dispatchHookChainFromHookRuntime({
+    eventName: 'PostToolUseFailure',
+    outcome: 'failed',
+    payload: {
+      ...hookInput,
+      hook_blocking_error_count: blockingHookCount,
+      hook_execution_skipped: !hasPostToolFailureHooks,
+    },
     signal,
-    timeoutMs,
     toolUseContext,
   })
 }
@@ -3807,12 +3971,36 @@ export async function* executeTaskCompletedHooks(
     team_name: teamName,
   }
 
-  yield* executeHooks({
+  let blockingHookCount = 0
+  let preventedContinuation = false
+
+  for await (const result of executeHooks({
     hookInput,
     toolUseID: randomUUID(),
     signal,
     timeoutMs,
     toolUseContext,
+  })) {
+    if (result.blockingError) {
+      blockingHookCount++
+    }
+    if (result.preventContinuation) {
+      preventedContinuation = true
+    }
+    yield result
+  }
+
+  await dispatchHookChainFromHookRuntime({
+    eventName: 'TaskCompleted',
+    outcome:
+      blockingHookCount > 0 || preventedContinuation ? 'failed' : 'success',
+    payload: {
+      ...hookInput,
+      hook_blocking_error_count: blockingHookCount,
+      hook_prevented_continuation: preventedContinuation,
+    },
+    signal,
+    toolUseContext,
   })
 }
 

src/utils/messages.ts

@@ -75,6 +75,7 @@ import type {
 import { isAdvisorBlock } from './advisor.js'
 import { isAgentSwarmsEnabled } from './agentSwarmsEnabled.js'
 import { count } from './array.js'
+import { isEnvTruthy } from './envUtils.js'
 import {
   type Attachment,
   type HookAttachment,
@@ -3666,6 +3667,9 @@ Read the team config to discover your teammates' names. Check the task list peri
       ])
     }
     case 'todo_reminder': {
+      if (isEnvTruthy(process.env.OPENCLAUDE_DISABLE_TOOL_REMINDERS)) {
+        return []
+      }
       const todoItems = attachment.content
         .map((todo, index) => `${index + 1}. [${todo.status}] ${todo.content}`)
         .join('\n')
@@ -3686,6 +3690,9 @@ Read the team config to discover your teammates' names. Check the task list peri
       if (!isTodoV2Enabled()) {
         return []
       }
+      if (isEnvTruthy(process.env.OPENCLAUDE_DISABLE_TOOL_REMINDERS)) {
+        return []
+      }
       const taskItems = attachment.content
         .map(task => `#${task.id}. [${task.status}] ${task.subject}`)
         .join('\n')

src/utils/model/model.openai-shim-providers.test.ts

@@ -1,7 +1,13 @@
-import { afterEach, beforeEach, expect, test } from 'bun:test'
+import { afterEach, beforeEach, expect, mock, test } from 'bun:test'
 
 import { saveGlobalConfig } from '../config.js'
-import { getUserSpecifiedModelSetting } from './model.js'
+import {
+  getDefaultHaikuModel,
+  getDefaultOpusModel,
+  getDefaultSonnetModel,
+  getSmallFastModel,
+  getUserSpecifiedModelSetting,
+} from './model.js'
 
 const SAVED_ENV = {
   CLAUDE_CODE_USE_OPENAI: process.env.CLAUDE_CODE_USE_OPENAI,
@@ -28,6 +34,11 @@ function restoreEnv(key: keyof typeof SAVED_ENV): void {
 }
 
 beforeEach(() => {
+  // Other test files (notably modelOptions.github.test.ts) install a
+  // persistent mock.module for './providers.js' that overrides getAPIProvider
+  // globally. Without mock.restore() here, those overrides bleed into this
+  // suite and the provider-kind branches we're testing become unreachable.
+  mock.restore()
   delete process.env.CLAUDE_CODE_USE_OPENAI
   delete process.env.CLAUDE_CODE_USE_GEMINI
   delete process.env.CLAUDE_CODE_USE_GITHUB
@@ -113,3 +124,76 @@ test('github provider still reads OPENAI_MODEL (regression guard)', () => {
   expect(model).toBe('github:copilot')
 })
 
+// ---------------------------------------------------------------------------
+// Default model helpers — must not fall through to claude-haiku-4-5 etc. for
+// OpenAI-shim providers whose endpoints don't speak Anthropic model names.
+// Hitting that fallthrough caused WebFetch to hang for 60s on MiniMax/Codex
+// because queryHaiku() shipped an unknown model id to the shim endpoint.
+// ---------------------------------------------------------------------------
+
+test('getSmallFastModel returns OPENAI_MODEL for MiniMax (regression: WebFetch hang)', () => {
+  process.env.MINIMAX_API_KEY = 'minimax-test'
+  process.env.OPENAI_MODEL = 'MiniMax-M2.5-highspeed'
+
+  expect(getSmallFastModel()).toBe('MiniMax-M2.5-highspeed')
+})
+
+test('getSmallFastModel returns OPENAI_MODEL for Codex (regression)', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'https://chatgpt.com/backend-api/codex'
+  process.env.OPENAI_MODEL = 'codexspark'
+  process.env.CODEX_API_KEY = 'codex-test'
+  process.env.CHATGPT_ACCOUNT_ID = 'acct_test'
+
+  expect(getSmallFastModel()).toBe('codexspark')
+})
+
+test('getSmallFastModel returns OPENAI_MODEL for NVIDIA NIM (regression)', () => {
+  process.env.NVIDIA_NIM = '1'
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'nvidia/llama-3.1-nemotron-70b-instruct'
+
+  expect(getSmallFastModel()).toBe('nvidia/llama-3.1-nemotron-70b-instruct')
+})
+
+test('getDefaultOpusModel returns OPENAI_MODEL for MiniMax', () => {
+  process.env.MINIMAX_API_KEY = 'minimax-test'
+  process.env.OPENAI_MODEL = 'MiniMax-M2.7'
+
+  expect(getDefaultOpusModel()).toBe('MiniMax-M2.7')
+})
+
+test('getDefaultSonnetModel returns OPENAI_MODEL for NVIDIA NIM', () => {
+  process.env.NVIDIA_NIM = '1'
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'nvidia/llama-3.1-nemotron-70b-instruct'
+
+  expect(getDefaultSonnetModel()).toBe('nvidia/llama-3.1-nemotron-70b-instruct')
+})
+
+test('getDefaultHaikuModel returns OPENAI_MODEL for MiniMax', () => {
+  process.env.MINIMAX_API_KEY = 'minimax-test'
+  process.env.OPENAI_MODEL = 'MiniMax-M2.5-highspeed'
+
+  expect(getDefaultHaikuModel()).toBe('MiniMax-M2.5-highspeed')
+})
+
+test('default helpers do not leak claude-* names to shim providers', () => {
+  // Umbrella guard: for each OpenAI-shim provider, none of the default-model
+  // helpers may return an Anthropic-branded model name. That was the source
+  // of the WebFetch 60s hang — MiniMax received "claude-haiku-4-5" and sat
+  // on the connection.
+  process.env.MINIMAX_API_KEY = 'minimax-test'
+  process.env.OPENAI_MODEL = 'MiniMax-M2.7'
+
+  for (const fn of [
+    getSmallFastModel,
+    getDefaultOpusModel,
+    getDefaultSonnetModel,
+    getDefaultHaikuModel,
+  ]) {
+    const model = fn()
+    expect(model.toLowerCase()).not.toContain('claude')
+  }
+})
+

src/utils/model/model.ts

@@ -52,10 +52,25 @@ export function getSmallFastModel(): ModelName {
   if (getAPIProvider() === 'openai') {
     return process.env.OPENAI_MODEL || 'gpt-4o-mini'
   }
+  // Codex provider — OPENAI_MODEL is always set for Codex profiles; only fall
+  // back to a codex-spark alias when an override env strips it.
+  if (getAPIProvider() === 'codex') {
+    return process.env.OPENAI_MODEL || 'codexspark'
+  }
   // For GitHub Copilot provider
   if (getAPIProvider() === 'github') {
     return process.env.OPENAI_MODEL || 'github:copilot'
   }
+  // NVIDIA NIM — OPENAI_MODEL carries the user's active NIM model; use a
+  // small Meta Llama variant as the conservative fallback.
+  if (getAPIProvider() === 'nvidia-nim') {
+    return process.env.OPENAI_MODEL || 'meta/llama-3.1-8b-instruct'
+  }
+  // MiniMax — OPENAI_MODEL carries the active MiniMax model; fall back to
+  // the fastest tier (M2.5-highspeed) when missing.
+  if (getAPIProvider() === 'minimax') {
+    return process.env.OPENAI_MODEL || 'MiniMax-M2.5-highspeed'
+  }
   return getDefaultHaikuModel()
 }
 
@@ -171,6 +186,14 @@ export function getDefaultOpusModel(): ModelName {
   if (getAPIProvider() === 'github') {
     return process.env.OPENAI_MODEL || 'github:copilot'
   }
+  // NVIDIA NIM
+  if (getAPIProvider() === 'nvidia-nim') {
+    return process.env.OPENAI_MODEL || 'nvidia/llama-3.1-nemotron-70b-instruct'
+  }
+  // MiniMax — flagship tier for "opus"-equivalent.
+  if (getAPIProvider() === 'minimax') {
+    return process.env.OPENAI_MODEL || 'MiniMax-M2.7'
+  }
   // 3P providers (Bedrock, Vertex, Foundry) — kept as a separate branch
   // even when values match, since 3P availability lags firstParty and
   // these will diverge again at the next model launch.
@@ -205,6 +228,14 @@ export function getDefaultSonnetModel(): ModelName {
   if (getAPIProvider() === 'github') {
     return process.env.OPENAI_MODEL || 'github:copilot'
   }
+  // NVIDIA NIM
+  if (getAPIProvider() === 'nvidia-nim') {
+    return process.env.OPENAI_MODEL || 'nvidia/llama-3.1-nemotron-70b-instruct'
+  }
+  // MiniMax — mid tier for "sonnet"-equivalent.
+  if (getAPIProvider() === 'minimax') {
+    return process.env.OPENAI_MODEL || 'MiniMax-M2.5'
+  }
   // Default to Sonnet 4.5 for 3P since they may not have 4.6 yet
   if (getAPIProvider() !== 'firstParty') {
     return getModelStrings().sonnet45
@@ -237,6 +268,14 @@ export function getDefaultHaikuModel(): ModelName {
   if (getAPIProvider() === 'gemini') {
     return process.env.GEMINI_MODEL || 'gemini-2.0-flash-lite'
   }
+  // NVIDIA NIM
+  if (getAPIProvider() === 'nvidia-nim') {
+    return process.env.OPENAI_MODEL || 'meta/llama-3.1-8b-instruct'
+  }
+  // MiniMax — fastest tier for "haiku"-equivalent.
+  if (getAPIProvider() === 'minimax') {
+    return process.env.OPENAI_MODEL || 'MiniMax-M2.5-highspeed'
+  }
 
   // Haiku 4.5 is available on all platforms (first-party, Foundry, Bedrock, Vertex)
   return getModelStrings().haiku45

src/utils/model/openaiContextWindows.ts

@@ -413,16 +413,51 @@ const OPENAI_MAX_OUTPUT_TOKENS: Record<string, number> = {
   'moonshot-v1-128k':          32_768,
 }
 
-function lookupByModel<T>(table: Record<string, T>, model: string): T | undefined {
+// External context-window overrides loaded once at startup.
+// Set CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS to a JSON object mapping model name
+// → context-window token count to add or override entries without editing
+// this file.  Example:
+//   CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS='{"my-corp/llm-v2":200000}'
+const OPENAI_EXTERNAL_CONTEXT_WINDOWS: Record<string, number> = (() => {
+  try {
+    const raw = process.env.CLAUDE_CODE_OPENAI_CONTEXT_WINDOWS
+    if (raw) {
+      const parsed = JSON.parse(raw)
+      if (typeof parsed === 'object' && parsed !== null) return parsed as Record<string, number>
+    }
+  } catch { /* ignore malformed JSON */ }
+  return {}
+})()
+
+// External max-output-token overrides.
+// Set CLAUDE_CODE_OPENAI_MAX_OUTPUT_TOKENS to a JSON object mapping model name
+// → max output token count.
+const OPENAI_EXTERNAL_MAX_OUTPUT_TOKENS: Record<string, number> = (() => {
+  try {
+    const raw = process.env.CLAUDE_CODE_OPENAI_MAX_OUTPUT_TOKENS
+    if (raw) {
+      const parsed = JSON.parse(raw)
+      if (typeof parsed === 'object' && parsed !== null) return parsed as Record<string, number>
+    }
+  } catch { /* ignore malformed JSON */ }
+  return {}
+})()
+
+function lookupByModel<T>(table: Record<string, T>, externalTable: Record<string, T>, model: string): T | undefined {
   // Try provider-qualified key first: "{OPENAI_MODEL}:{model}" so that
   // e.g. "github:copilot:claude-haiku-4.5" can have different limits than
   // a bare "claude-haiku-4.5" served by another provider.
   const providerModel = process.env.OPENAI_MODEL?.trim()
   if (providerModel && providerModel !== model) {
     const qualified = `${providerModel}:${model}`
+    // External table takes precedence over the built-in table.
+    const externalQualified = lookupByKey(externalTable, qualified)
+    if (externalQualified !== undefined) return externalQualified
     const qualifiedResult = lookupByKey(table, qualified)
     if (qualifiedResult !== undefined) return qualifiedResult
   }
+  const externalResult = lookupByKey(externalTable, model)
+  if (externalResult !== undefined) return externalResult
   return lookupByKey(table, model)
 }
 
@@ -446,7 +481,7 @@ function lookupByKey<T>(table: Record<string, T>, model: string): T | undefined
  * "gpt-4o-2024-11-20" resolve to the base "gpt-4o" entry.
  */
 export function getOpenAIContextWindow(model: string): number | undefined {
-  return lookupByModel(OPENAI_CONTEXT_WINDOWS, model)
+  return lookupByModel(OPENAI_CONTEXT_WINDOWS, OPENAI_EXTERNAL_CONTEXT_WINDOWS, model)
 }
 
 /**
@@ -454,5 +489,5 @@ export function getOpenAIContextWindow(model: string): number | undefined {
  * Returns undefined if the model is not in the table.
  */
 export function getOpenAIMaxOutputTokens(model: string): number | undefined {
-  return lookupByModel(OPENAI_MAX_OUTPUT_TOKENS, model)
+  return lookupByModel(OPENAI_MAX_OUTPUT_TOKENS, OPENAI_EXTERNAL_MAX_OUTPUT_TOKENS, model)
 }

src/utils/model/providers.ts

@@ -19,7 +19,12 @@ export function getAPIProvider(): APIProvider {
   if (isEnvTruthy(process.env.NVIDIA_NIM)) {
     return 'nvidia-nim'
   }
-  if (isEnvTruthy(process.env.MINIMAX_API_KEY)) {
+  // MiniMax is signalled by a real API key, not a '1'/'true' flag. Using
+  // isEnvTruthy() here silently treated every MiniMax user as 'firstParty'
+  // (or 'openai' once they set CLAUDE_CODE_USE_OPENAI via the profile),
+  // making every provider-kind-specific branch for 'minimax' elsewhere in
+  // the codebase unreachable. Presence check is the correct signal.
+  if (typeof process.env.MINIMAX_API_KEY === 'string' && process.env.MINIMAX_API_KEY.trim() !== '') {
     return 'minimax'
   }
   return isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)

src/utils/requestLogging.test.ts

@@ -0,0 +1,86 @@
+import { describe, expect, it, beforeEach } from 'bun:test'
+import {
+  createCorrelationId,
+  logApiCallStart,
+  logApiCallEnd,
+} from './requestLogging.js'
+
+describe('requestLogging', () => {
+  describe('createCorrelationId', () => {
+    it('returns a non-empty string', () => {
+      const id = createCorrelationId()
+      expect(id).toBeTruthy()
+      expect(typeof id).toBe('string')
+    })
+
+    it('returns unique IDs', () => {
+      const id1 = createCorrelationId()
+      const id2 = createCorrelationId()
+      expect(id1).not.toBe(id2)
+    })
+  })
+
+  describe('logApiCallStart', () => {
+    it('returns correlation ID and start time', () => {
+      const result = logApiCallStart('openai', 'gpt-4o')
+      expect(result.correlationId).toBeTruthy()
+      expect(result.startTime).toBeGreaterThan(0)
+    })
+
+    it('logs without throwing', () => {
+      expect(() => logApiCallStart('ollama', 'llama3')).not.toThrow()
+    })
+  })
+
+  describe('logApiCallEnd', () => {
+    it('logs success without throwing', () => {
+      const { correlationId, startTime } = logApiCallStart('openai', 'gpt-4o')
+      expect(() =>
+        logApiCallEnd(
+          correlationId,
+          startTime,
+          'gpt-4o',
+          'success',
+          100,
+          50,
+          false,
+        ),
+      ).not.toThrow()
+    })
+
+    it('logs error without throwing', () => {
+      const { correlationId, startTime } = logApiCallStart('openai', 'gpt-4o')
+      expect(() =>
+        logApiCallEnd(
+          correlationId,
+          startTime,
+          'gpt-4o',
+          'error',
+          0,
+          0,
+          false,
+          undefined,
+          undefined,
+          'Network error',
+        ),
+      ).not.toThrow()
+    })
+
+    it('logs with all parameters without throwing', () => {
+      const { correlationId, startTime } = logApiCallStart('openai', 'gpt-4o')
+      expect(() =>
+        logApiCallEnd(
+          correlationId,
+          startTime,
+          'gpt-4o',
+          'success',
+          100,
+          50,
+          true,
+          'error message',
+          { provider: 'openai' },
+        ),
+      ).not.toThrow()
+    })
+  })
+})

src/utils/requestLogging.ts

@@ -0,0 +1,89 @@
+/**
+ * Structured Request Logging
+ * 
+ * Uses existing logForDebugging for structured logging.
+ */
+
+import { randomUUID } from 'crypto'
+import { logForDebugging } from './debug.js'
+
+export interface RequestLog {
+  correlationId: string
+  timestamp: number
+  provider: string
+  model: string
+  duration: number
+  status: 'success' | 'error'
+  tokensIn: number
+  tokensOut: number
+  error?: string
+  streaming: boolean
+}
+
+export function createCorrelationId(): string {
+  return randomUUID()
+}
+
+export function logApiCallStart(
+  provider: string,
+  model: string,
+): { correlationId: string; startTime: number } {
+  const correlationId = createCorrelationId()
+  const startTime = Date.now()
+
+  logForDebugging(
+    JSON.stringify({
+      type: 'api_call_start',
+      correlationId,
+      provider,
+      model,
+      timestamp: startTime,
+    }),
+    { level: 'debug' },
+  )
+
+  return { correlationId, startTime }
+}
+
+export function logApiCallEnd(
+  correlationId: string,
+  startTime: number,
+  model: string,
+  status: 'success' | 'error',
+  tokensIn: number,
+  tokensOut: number,
+  streaming: boolean,
+  firstTokenMs?: number,
+  totalChunks?: number,
+  error?: string,
+): void {
+  const duration = Date.now() - startTime
+
+  const logData: Record<string, unknown> = {
+    type: status === 'error' ? 'api_call_error' : 'api_call_end',
+    correlationId,
+    model,
+    duration_ms: duration,
+    status,
+    tokens_in: tokensIn,
+    tokens_out: tokensOut,
+    streaming,
+  }
+
+  if (firstTokenMs !== undefined) {
+    logData.first_token_ms = firstTokenMs
+  }
+
+  if (totalChunks !== undefined) {
+    logData.total_chunks = totalChunks
+  }
+
+  if (error) {
+    logData.error = error
+  }
+
+  logForDebugging(
+    JSON.stringify(logData),
+    { level: status === 'error' ? 'error' : 'debug' },
+  )
+}

src/utils/streamingOptimizer.test.ts

@@ -0,0 +1,61 @@
+import { describe, expect, it, beforeEach } from 'bun:test'
+import {
+  createStreamState,
+  processStreamChunk,
+  flushStreamBuffer,
+  getStreamStats,
+} from './streamingOptimizer.js'
+
+describe('streamingOptimizer', () => {
+  let state: ReturnType<typeof createStreamState>
+
+  beforeEach(() => {
+    state = createStreamState()
+  })
+
+  describe('createStreamState', () => {
+    it('creates initial state with zero counts', () => {
+      expect(state.chunkCount).toBe(0)
+      expect(state.firstTokenTime).toBeNull()
+      expect(state.startTime).toBeGreaterThan(0)
+    })
+  })
+
+  describe('processStreamChunk', () => {
+    it('tracks first token time on first chunk', () => {
+      processStreamChunk(state, 'hello')
+      expect(state.firstTokenTime).not.toBeNull()
+      expect(state.chunkCount).toBe(1)
+    })
+
+    it('increments chunk count', () => {
+      processStreamChunk(state, 'chunk1')
+      processStreamChunk(state, 'chunk2')
+      expect(state.chunkCount).toBe(2)
+    })
+  })
+
+  describe('getStreamStats', () => {
+    it('returns zero values for empty stream', () => {
+      const stats = getStreamStats(state)
+      expect(stats.totalChunks).toBe(0)
+      expect(stats.firstTokenMs).toBeNull()
+      expect(stats.durationMs).toBeGreaterThanOrEqual(0)
+    })
+
+    it('returns correct stats after processing chunks', () => {
+      processStreamChunk(state, 'test')
+      const stats = getStreamStats(state)
+      expect(stats.totalChunks).toBe(1)
+      expect(stats.firstTokenMs).toBeGreaterThanOrEqual(0)
+      expect(stats.durationMs).toBeGreaterThanOrEqual(0)
+    })
+  })
+
+  describe('flushStreamBuffer', () => {
+    it('returns empty string (no-op)', () => {
+      const result = flushStreamBuffer(state)
+      expect(result).toBe('')
+    })
+  })
+})

src/utils/streamingOptimizer.ts

@@ -0,0 +1,51 @@
+/**
+ * Streaming Stats Tracker
+ * 
+ * Observational stats tracking for streaming responses.
+ * No buffering - purely tracks metrics for monitoring.
+ */
+
+export interface StreamStats {
+  totalChunks: number
+  firstTokenMs: number | null
+  durationMs: number
+}
+
+export interface StreamState {
+  chunkCount: number
+  firstTokenTime: number | null
+  startTime: number
+}
+
+export function createStreamState(): StreamState {
+  return {
+    chunkCount: 0,
+    firstTokenTime: null,
+    startTime: Date.now(),
+  }
+}
+
+export function processStreamChunk(state: StreamState, _chunk: string): void {
+  if (state.firstTokenTime === null) {
+    state.firstTokenTime = Date.now()
+  }
+  state.chunkCount++
+}
+
+export function flushStreamBuffer(_state: StreamState): string {
+  return '' // No-op - kept for API compatibility
+}
+
+export function getStreamStats(state: StreamState): StreamStats {
+  const now = Date.now()
+  const firstTokenMs = state.firstTokenTime
+    ? now - state.firstTokenTime
+    : null
+  const durationMs = now - state.startTime
+
+  return {
+    totalChunks: state.chunkCount,
+    firstTokenMs,
+    durationMs,
+  }
+}