fix(shims): strip x-anthropic-billing-header block before forwarding system prompt (Gitlawb#1019)

`getAttributionHeader()` (src/constants/system.ts) builds an
`x-anthropic-billing-header: cc_version=...; cc_entrypoint=...; ...` line
that gets prepended to the system-prompt block array in
src/services/api/claude.ts:1390-1401. The Anthropic API path needs it
(server _parse_cc_header consumes it), but the OpenAI / Codex shims
joined every text block straight into the outbound `system` /
`instructions` payload, so non-Anthropic providers received an
Anthropic-only billing string in their prompt — token waste, plus a
per-build fingerprint that churns local-model KV cache and any
upstream prompt cache (the slowdown unsloth flagged for Claude Code).

Fix the two `convertSystemPrompt` helpers (openaiShim.ts:241,
codexShim.ts:124) to drop blocks whose text starts with
`x-anthropic-billing-header`. Anthropic-bound traffic is unaffected —
the block is built into Anthropic-shaped requests directly and never
flows through these helpers.

Tests:
- openaiShim.test.ts: e2e capturedBody assertions on chat-completions
  + responses-API paths confirm the line is absent and the rest of the
  system prompt survives.
- codexShim.test.ts: convertSystemPrompt is now exported (pure helper)
  and unit-tested for array + only-attribution + plain-string cases.

Closes Gitlawb#607.

Author: 0xfandom

src/services/api/codexShim.test.ts

@@ -6,6 +6,7 @@ import {
   codexStreamToAnthropic,
   convertAnthropicMessagesToResponsesInput,
   convertCodexResponseToAnthropicMessage,
+  convertSystemPrompt,
   convertToolsToResponsesTools,
 } from './codexShim.js'
 import { __test as webSearchToolTest } from '../../tools/WebSearchTool/WebSearchTool.js'
@@ -883,3 +884,40 @@ describe('Codex request translation', () => {
     )
   })
 })
+
+describe('convertSystemPrompt', () => {
+  test('strips Anthropic attribution header block from text-block array (#607)', () => {
+    const result = convertSystemPrompt([
+      {
+        type: 'text',
+        text:
+          'x-anthropic-billing-header: cc_version=0.8.0.abc123; ' +
+          'cc_entrypoint=cli;',
+      },
+      { type: 'text', text: 'You are Claude Code.' },
+      { type: 'text', text: 'Project context: bun + react.' },
+    ])
+
+    expect(result).not.toContain('x-anthropic-billing-header')
+    expect(result).not.toContain('cc_version=')
+    expect(result).toContain('You are Claude Code.')
+    expect(result).toContain('Project context: bun + react.')
+  })
+
+  test('returns empty string when only the attribution block is present', () => {
+    const result = convertSystemPrompt([
+      {
+        type: 'text',
+        text: 'x-anthropic-billing-header: cc_version=0.8.0.abc;',
+      },
+    ])
+
+    expect(result).toBe('')
+  })
+
+  test('passes plain string system prompts through untouched', () => {
+    expect(convertSystemPrompt('You are Claude Code.')).toBe(
+      'You are Claude Code.',
+    )
+  })
+})

src/services/api/codexShim.ts

@@ -121,14 +121,18 @@ function normalizeToolUseId(toolUseId: string | undefined): {
   }
 }
 
-function convertSystemPrompt(system: unknown): string {
+export function convertSystemPrompt(system: unknown): string {
   if (!system) return ''
   if (typeof system === 'string') return system
   if (Array.isArray(system)) {
     return system
       .map((block: { type?: string; text?: string }) =>
         block.type === 'text' ? (block.text ?? '') : '',
       )
+      // Drop the Anthropic billing/attribution block — Codex's Responses API
+      // doesn't parse it and the per-build fingerprint just churns the
+      // upstream prompt cache.
+      .filter(text => !text.startsWith('x-anthropic-billing-header'))
       .join('\n\n')
   }
   return String(system)

src/services/api/openaiShim.test.ts

@@ -4923,3 +4923,99 @@ test('Z.AI: thinking mode enabled when requested', async () => {
   expect(requestBody?.max_completion_tokens).toBeUndefined()
   expect(requestBody?.max_tokens).toBe(1024)
 })
+
+test('strips Anthropic attribution header block from chat-completions system prompt (#607)', async () => {
+  let capturedBody: Record<string, unknown> | undefined
+
+  globalThis.fetch = (async (_input, init) => {
+    capturedBody = JSON.parse(String(init?.body)) as Record<string, unknown>
+
+    return new Response(
+      JSON.stringify({
+        id: 'chatcmpl-1',
+        model: 'gpt-4o',
+        choices: [
+          {
+            message: { role: 'assistant', content: 'ok' },
+            finish_reason: 'stop',
+          },
+        ],
+        usage: { prompt_tokens: 8, completion_tokens: 3, total_tokens: 11 },
+      }),
+      { headers: { 'Content-Type': 'application/json' } },
+    )
+  }) as FetchType
+
+  const client = createOpenAIShimClient({}) as OpenAIShimClient
+
+  await client.beta.messages.create({
+    model: 'gpt-4o',
+    system: [
+      {
+        type: 'text',
+        text:
+          'x-anthropic-billing-header: cc_version=0.8.0.abc123; ' +
+          'cc_entrypoint=cli;',
+      },
+      { type: 'text', text: 'You are Claude Code, helpful assistant.' },
+      { type: 'text', text: 'Project context: bun + react.' },
+    ],
+    messages: [{ role: 'user', content: 'hello' }],
+    max_tokens: 64,
+    stream: false,
+  })
+
+  const messages = capturedBody?.messages as Array<{ role: string; content: string }>
+  const sysMsg = messages.find(m => m.role === 'system')
+  expect(sysMsg).toBeDefined()
+  expect(sysMsg?.content).not.toContain('x-anthropic-billing-header')
+  expect(sysMsg?.content).not.toContain('cc_version=')
+  expect(sysMsg?.content).toContain('You are Claude Code, helpful assistant.')
+  expect(sysMsg?.content).toContain('Project context: bun + react.')
+})
+
+test('strips Anthropic attribution header block from responses-API instructions (#607)', async () => {
+  process.env.OPENAI_API_FORMAT = 'responses'
+  let capturedBody: Record<string, unknown> | undefined
+
+  globalThis.fetch = (async (_input, init) => {
+    capturedBody = JSON.parse(String(init?.body)) as Record<string, unknown>
+
+    return new Response(
+      JSON.stringify({
+        id: 'resp-1',
+        model: 'gpt-5.4',
+        output: [
+          {
+            type: 'message',
+            role: 'assistant',
+            content: [{ type: 'output_text', text: 'ok' }],
+          },
+        ],
+        usage: { input_tokens: 8, output_tokens: 3, total_tokens: 11 },
+      }),
+      { headers: { 'Content-Type': 'application/json' } },
+    )
+  }) as FetchType
+
+  const client = createOpenAIShimClient({ defaultHeaders: {} }) as OpenAIShimClient
+
+  await client.beta.messages.create({
+    model: 'gpt-5.4',
+    system: [
+      {
+        type: 'text',
+        text: 'x-anthropic-billing-header: cc_version=0.8.0.abc123; cc_entrypoint=cli;',
+      },
+      { type: 'text', text: 'You are Claude Code.' },
+    ],
+    messages: [{ role: 'user', content: 'hello' }],
+    max_tokens: 64,
+    stream: false,
+  })
+
+  const instructions = capturedBody?.instructions as string
+  expect(instructions).not.toContain('x-anthropic-billing-header')
+  expect(instructions).not.toContain('cc_version=')
+  expect(instructions).toContain('You are Claude Code.')
+})

src/services/api/openaiShim.ts

@@ -248,6 +248,11 @@ function convertSystemPrompt(
       .map((block: { type?: string; text?: string }) =>
         block.type === 'text' ? block.text ?? '' : '',
       )
+      // Drop the Anthropic billing/attribution block — it's only meaningful to
+      // Anthropic's `_parse_cc_header` and is dead weight (plus a churning
+      // per-build fingerprint that busts prefix KV cache) for OpenAI-compat
+      // providers like local Ollama / llama.cpp / Codex pass-throughs.
+      .filter(text => !text.startsWith('x-anthropic-billing-header'))
       .join('\n\n')
   }
   return String(system)