fix: disable experimental API betas by default, reduce side query token usage, standardize Headers type (#281)

* fix: disable experimental API betas by default to prevent 500 errors

Tool search (defer_loading), global cache scope, and context management
betas require internal Anthropic server-side support. External accounts
receive 500 Internal Server Error when these are sent.

Set CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=true by default in the CLI
entrypoint. Users with internal access can opt back in with =false.

Also includes: cache key stability fixes (Sonnet 1M latch, system-before-
messages key ordering, resume fingerprint isMeta skip), sideQuery default
cleanup, and /dream command.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* refactor: standardize API headers to Headers type and enable tengu feature flags by default

* fix: address PR review — dream lock, MCP betas guard, redundant Partial

- Call recordConsolidation() programmatically in /dream instead of
  delegating to model prompt (unreliable)
- Add CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS guard to MCP entrypoint
  (was only in CLI entrypoint, causing 500s in MCP server mode)
- Remove redundant ? markers from SecretValueSource Partial<{}> type

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/utils/claudeInChrome/mcpServer.ts

@@ -181,8 +181,6 @@ export function createChromeContext(
         usage?: { input_tokens: number; output_tokens: number }
       }> => {
         // sideQuery handles OAuth attribution fingerprint, proxy, model betas.
-        // skipSystemPromptPrefix: the lightning prompt is complete on its own;
-        // the CLI prefix would dilute the batching instructions.
         // tools: [] is load-bearing — without it Sonnet emits
         // <function_calls> XML before the text commands. Original
         // lightning-harness.js (apps repo) does the same.
@@ -193,7 +191,6 @@ export function createChromeContext(
           max_tokens: req.max_tokens,
           stop_sequences: req.stop_sequences,
           signal: req.signal,
-          skipSystemPromptPrefix: true,
           tools: [],
           querySource: 'chrome_mcp',
         })

src/utils/fingerprint.ts

@@ -16,7 +16,13 @@ export const FINGERPRINT_SALT = '59cf53e54c78'
 export function extractFirstMessageText(
   messages: (UserMessage | AssistantMessage)[],
 ): string {
-  const firstUserMessage = messages.find(msg => msg.type === 'user')
+  // Skip isMeta messages (system-injected attachments) so the fingerprint
+  // reflects the actual user input. On --resume, reorderAttachmentsForAPI
+  // can bubble meta messages before the real first user message, changing
+  // the fingerprint and breaking cache.
+  const firstUserMessage =
+    messages.find(msg => msg.type === 'user' && !msg.isMeta) ??
+    messages.find(msg => msg.type === 'user')
   if (!firstUserMessage) {
     return ''
   }

src/utils/mcpInstructionsDelta.ts

@@ -39,7 +39,7 @@ export function isMcpInstructionsDeltaEnabled(): boolean {
   if (isEnvDefinedFalsy(process.env.CLAUDE_CODE_MCP_INSTR_DELTA)) return false
   return (
     process.env.USER_TYPE === 'ant' ||
-    getFeatureValue_CACHED_MAY_BE_STALE('tengu_basalt_3kr', false)
+    getFeatureValue_CACHED_MAY_BE_STALE('tengu_basalt_3kr', true)
   )
 }
 

src/utils/permissions/yoloClassifier.ts

@@ -780,7 +780,6 @@ async function classifyYoloActionXml(
         model,
         max_tokens: (mode === 'fast' ? 256 : 64) + thinkingPadding,
         system: systemBlocks,
-        skipSystemPromptPrefix: true,
         temperature: 0,
         thinking: disableThinking,
         messages: [
@@ -867,7 +866,6 @@ async function classifyYoloActionXml(
       model,
       max_tokens: 4096 + thinkingPadding,
       system: systemBlocks,
-      skipSystemPromptPrefix: true,
       temperature: 0,
       thinking: disableThinking,
       messages: [
@@ -1141,7 +1139,6 @@ export async function classifyYoloAction(
           cache_control: getCacheControl({ querySource: 'auto_mode' }),
         },
       ],
-      skipSystemPromptPrefix: true,
       temperature: 0,
       thinking: disableThinking,
       messages: [

src/utils/sideQuery.ts

@@ -51,7 +51,7 @@ export type SideQueryOptions = {
   maxRetries?: number
   /** Abort signal */
   signal?: AbortSignal
-  /** Skip CLI system prompt prefix (keeps attribution header for OAuth). For internal classifiers that provide their own prompt. */
+  /** Skip CLI system prompt prefix (keeps attribution header for OAuth). Default true — side queries are internal classifiers with their own prompt. Set false only for queries that need the full "You are Claude Code…" prefix. */
   skipSystemPromptPrefix?: boolean
   /** Temperature override */
   temperature?: number
@@ -115,7 +115,7 @@ export async function sideQuery(opts: SideQueryOptions): Promise<BetaMessage> {
     max_tokens = 1024,
     maxRetries = 2,
     signal,
-    skipSystemPromptPrefix,
+    skipSystemPromptPrefix = true,
     temperature,
     thinking,
     stop_sequences,