Commit Graph

1376 Commits

Author SHA1 Message Date
Luis Pater
0e6bb076e9 fix(translator): comment out service_tier removal from OpenAI response processing 2026-03-05 22:49:38 +08:00
Luis Pater
ac135fc7cb Fixed: #1815
**test(executor): add unit tests for prompt cache key generation in OpenAI `cacheHelper`**
2026-03-05 22:49:23 +08:00
Luis Pater
4e1d09809d Fixed: #1741
fix(translator): handle tool name mappings and improve tool call handling in OpenAI and Claude integrations
2026-03-05 22:24:50 +08:00
Luis Pater
5850492a93 Fixed: #1548
test(translator): add unit tests for fallback logic in `ConvertCodexResponseToOpenAI` model assignment
2026-03-05 12:11:54 +08:00
Luis Pater
fdbd4041ca Fixed: #1531
fix(gemini): add `deprecated` to unsupported schema keywords

Add `deprecated` to the list of unsupported schema metadata fields in Gemini and update tests to verify its removal.
2026-03-05 11:48:15 +08:00
Luis Pater
ebef1fae2a Merge pull request #1511 from stondy0103/fix/responses-nullable-type-array
fix(translator): fix nullable type arrays breaking Gemini/Antigravity API
2026-03-05 11:30:09 +08:00
Luis Pater
4bbeb92e9a Fixed: #1135
**test(translator): add tests for `tool_choice` handling in Claude request conversions**
2026-03-04 22:28:26 +08:00
Luis Pater
b436dad8bc Merge pull request #1822 from sususu98/fix/strip-defer-loading
fix(translator): strip defer_loading from Claude tool declarations in Codex and Gemini translators
2026-03-04 20:49:48 +08:00
Luis Pater
0468bde0d6 Merge branch 'dev' into fix/antigravity-adaptive-effort 2026-03-04 20:48:26 +08:00
hkfires
48ffc4dee7 feat(config): support excluded vertex models in config 2026-03-04 18:47:42 +08:00
sususu98
d26ad8224d fix(translator): strip defer_loading from Claude tool declarations in Codex and Gemini translators
Claude's Tool Search feature (advanced-tool-use-2025-11-20 beta) adds
defer_loading field to tool definitions. When proxying Claude requests
to Codex or Gemini, this unknown field causes 400 errors upstream.

Strip defer_loading (and cache_control where missing) in all three
Claude-to-upstream translation paths:
- codex/claude: defer_loading + cache_control
- gemini-cli/claude: defer_loading
- gemini/claude: defer_loading

Fixes #1725, Fixes #1375
2026-03-04 14:21:30 +08:00
hkfires
5c84d69d42 feat(translator): map output_config.effort to adaptive thinking level in antigravity 2026-03-04 13:11:07 +08:00
sususu98
527e4b7f26 fix(antigravity): pass through adaptive thinking effort level instead of always mapping to high 2026-03-04 10:12:45 +08:00
Luis Pater
b48485b42b Fixed: #822
**fix(auth): normalize ID casing on Windows to prevent duplicate entries due to case-insensitive paths**
2026-03-04 02:31:20 +08:00
Luis Pater
9f95b31158 **fix(translator): enhance handling of mixed output content in Claude requests** 2026-03-03 21:49:41 +08:00
Luis Pater
5da07eae4c Merge pull request #1805 from router-for-me/thinking
Add adaptive thinking support for Claude models
2026-03-03 20:31:31 +08:00
hkfires
835ae178d4 feat(thinking): rename isBudgetBasedProvider to isBudgetCapableProvider and update logic for provider checks 2026-03-03 19:49:51 +08:00
hkfires
c80ab8bf0d feat(thinking): improve provider family checks and clamp unsupported levels 2026-03-03 19:05:15 +08:00
hkfires
ce87714ef1 feat(thinking): normalize effort levels in adaptive thinking requests to prevent validation errors 2026-03-03 15:10:47 +08:00
hkfires
0452b869e8 feat(thinking): add HasLevel and MapToClaudeEffort functions for adaptive thinking support 2026-03-03 14:16:36 +08:00
hkfires
d2e5857b82 feat(thinking): enhance adaptive thinking support across models and update test cases 2026-03-03 13:00:24 +08:00
Luis Pater
f9b005f21f Fixed: #1799
**test(auth): add tests for auth file deletion logic with manager and fallback scenarios**
2026-03-03 09:37:24 +08:00
hkfires
c44793789b feat(thinking): add adaptive thinking support for Claude models
Add support for Claude's "adaptive" and "auto" thinking modes using `output_config.effort`. Introduce support for new effort level "max" in adaptive thinking. Update thinking logic, validate model capabilities, and extend converters and handling to ensure compatibility with adaptive modes. Adjust static model data with supported levels and refine handling across translators and executors.
2026-03-03 09:05:31 +08:00
hkfires
9229708b6c revert(executor): re-apply PR #1735 antigravity changes with cleanup 2026-03-02 19:30:32 +08:00
hkfires
914db94e79 refactor(headers): streamline User-Agent handling and introduce GeminiCLI versioning 2026-03-02 13:04:30 +08:00
hkfires
660bd7eff5 refactor(config): remove oauth-model-alias migration logic and related tests 2026-03-02 13:02:15 +08:00
hkfires
b907d21851 revert(executor): revert antigravity_executor.go changes from PR #1735 2026-03-02 12:54:15 +08:00
Luis Pater
d6cc976d1f chore(executor): remove unused header scrubbing function 2026-03-02 03:40:54 +08:00
Luis Pater
8aa2cce8c5 Merge PR #1735 into dev with conflict resolution and fixes 2026-03-02 03:22:51 +08:00
Luis Pater
77b42c6165 fix(claude): handle X-CPA-CLAUDE-1M header and ensure proper beta merging logic 2026-03-01 21:39:33 +08:00
Luis Pater
1cbc4834e1 Merge pull request #1771 from edlsh/fix/claude-cache-control-1769
Fix Claude OAuth cache_control regressions and gzip error decoding
2026-03-01 20:17:22 +08:00
edlsh
76aa917882 Optimize cache-control JSON mutations in Claude executor 2026-02-28 22:47:04 -05:00
edlsh
6ac9b31e4e Handle compressed error decode failures safely 2026-02-28 22:43:46 -05:00
edlsh
0ad3e8457f Clarify cloaking system block cache-control comments 2026-02-28 22:34:14 -05:00
edlsh
444a47ae63 Fix Claude cache-control guardrails and gzip error decoding 2026-02-28 22:32:33 -05:00
Luis Pater
725f4fdff4 Merge pull request #1768 from router-for-me/claude
fix(translator): handle Claude thinking type "auto" like adaptive
2026-03-01 11:03:13 +08:00
hkfires
b148820c35 fix(translator): handle Claude thinking type "auto" like adaptive 2026-03-01 10:30:19 +08:00
hkfires
134f41496d fix(antigravity): update model configurations and add new models for Antigravity 2026-03-01 10:05:29 +08:00
Luis Pater
1ae994b4aa fix(antigravity): adjust thinkingBudget default to 64000 and update model definitions for Claude 2026-03-01 09:39:39 +08:00
Luis Pater
cc1d8f6629 Fixed: #1747
feat(auth): add configurable max-retry-credentials for finer control over cross-credential retries
2026-03-01 02:42:36 +08:00
Luis Pater
5446cd2b02 Merge pull request #1761 from margbug01/fix/thinking-chain-display
fix: support thinking.type=auto from Amp client and decouple thinking translation from unsigned history
2026-03-01 02:30:42 +08:00
margbug01
8de0885b7d fix: support thinking.type="auto" from Amp client for Antigravity Claude models
## Problem

When using Antigravity Claude models through CLIProxyAPI, the thinking
chain (reasoning content) does not display in the Amp client.

## Root Cause

The Amp client sends `thinking: {"type": "auto"}` in its requests,
but `ConvertClaudeRequestToAntigravity` only handled `"enabled"` and
`"adaptive"` types in its switch statement. The `"auto"` type was
silently ignored, resulting in no `thinkingConfig` being set in the
translated Gemini request. Without `thinkingConfig`, the Antigravity
API returns responses without any thinking content.

Additionally, the Antigravity API for Claude models does not support
`thinkingBudget: -1` (auto mode sentinel). It requires a concrete
positive budget value. The fix uses 128000 as the budget for "auto"
mode, which `ApplyThinking` will then normalize to stay within the
model's actual limits (e.g., capped to `maxOutputTokens - 1`).

## Changes

### internal/translator/antigravity/claude/antigravity_claude_request.go

1. **Add "auto" case** to the thinking type switch statement.
   Sets `thinkingBudget: 128000` and `includeThoughts: true`.
   The budget is subsequently normalized by `ApplyThinking` based
   on model-specific limits.

2. **Add "auto" to hasThinking check** so that interleaved thinking
   hints are injected for tool-use scenarios when Amp sends
   `thinking.type="auto"`.

### internal/registry/model_definitions_static_data.go

3. **Add Thinking configuration** for `claude-sonnet-4-6`,
   `claude-sonnet-4-5`, and `claude-opus-4-6` in
   `GetAntigravityModelConfig()` -- these were previously missing,
   causing `ApplyThinking` to skip thinking config entirely.

## Testing

- Deployed to Railway test instance (cpa-thinking-test)
- Verified via debug logging that:
  - Amp sends `thinking: {"type": "auto"}`
  - CPA now translates this to `thinkingConfig: {thinkingBudget: 128000, includeThoughts: true}`
  - `ApplyThinking` normalizes the budget to model-specific limits
  - Antigravity API receives the correct thinkingConfig

Amp-Thread-ID: https://ampcode.com/threads/T-019ca511-710d-776d-a07c-4b750f871a93
Co-authored-by: Amp <amp@ampcode.com>
2026-03-01 02:18:43 +08:00
Luis Pater
a6ce5f36e6 Fixed: #1758
fix(codex): filter billing headers from system result text and update template logic
2026-03-01 01:45:35 +08:00
Luis Pater
e73cf42e28 Merge pull request #1750 from tpm2dot0/fix/claude-code-request-fingerprint-alignment
fix(cloak): align outgoing requests with real Claude Code 2.1.63
2026-03-01 01:27:28 +08:00
exe.dev user
b45343e812 fix(cloak): align outgoing requests with real Claude Code 2.1.63 fingerprint
Captured and compared outgoing requests from CLIProxyAPI against real
Claude Code 2.1.63 and fixed all detectable differences:

Headers:
- Update anthropic-beta to match 2.1.63: replace fine-grained-tool-streaming
  and prompt-caching-2024-07-31 with context-management-2025-06-27 and
  prompt-caching-scope-2026-01-05
- Remove X-Stainless-Helper-Method header (real Claude Code does not send it)
- Update default User-Agent from "claude-cli/2.1.44 (external, sdk-cli)" to
  "claude-cli/2.1.63 (external, cli)"
- Force Claude Code User-Agent for non-Claude clients to avoid leaking
  real client identity (e.g. curl, OpenAI SDKs) during cloaking

Body:
- Inject x-anthropic-billing-header as system[0] (matches real format)
- Change system prompt identifier from "You are Claude Code..." to
  "You are a Claude agent, built on Anthropic's Claude Agent SDK."
- Add cache_control with ttl:"1h" to match real request format
- Fix user_id format: user_[64hex]_account_[uuid]_session_[uuid]
  (was missing account UUID)
- Disable tool name prefix (set claudeToolPrefix to empty string)

TLS:
- Switch utls fingerprint from HelloFirefox_Auto to HelloChrome_Auto
  (closer to Node.js/OpenSSL used by real Claude Code)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-28 09:19:06 +00:00
Luis Pater
8599b1560e Fixed: #1716
feat(kimi): add support for explicit disabled thinking and reasoning effort handling
2026-02-28 05:29:07 +08:00
Luis Pater
8bde8c37c0 Fixed: #1711
fix(server): use resolved log directory for request logger initialization and test fallback logic
2026-02-28 05:21:01 +08:00
maplelove
68dd2bfe82 fix(translator): allow passthrough of custom generationConfig for all Gemini-like providers 2026-02-27 17:13:42 +08:00
Luis Pater
41b1cf2273 Merge pull request #1734 from huangusaki/main
feat(registry): add gemini-3.1-flash-image support
2026-02-27 16:12:05 +08:00
maplelove
2baf35b3ef fix(executor): bump antigravity UA to 1.19.6 and align image_gen payload 2026-02-27 14:09:37 +08:00