Dev tama by el-pablos · Pull Request #6 · prassaaa/copilot-api

el-pablos · 2026-03-23T23:42:14Z

inject xhigh thinking gpt5.x & claude 4.5 model

el-pablos <yeteprem.end23juni@gmail.com>

- Add sessionId and userId optional fields to RequestContext - Add generateTraceId() for creating unique trace IDs - Add runWithContext() helper for running code with context - Add getTraceId() function (alias getCurrentTraceId as deprecated) el-pablos <yeteprem.end23juni@gmail.com>

…Delta dan ResponseMessage el-pablos <yeteprem.end23juni@gmail.com>

el-pablos <yeteprem.end23juni@gmail.com>

- Test untuk PATHS constants (APP_DIR, GITHUB_TOKEN_PATH) - Test untuk ensurePaths function (directory creation, file permissions) - Test untuk XDG Base Directory spec compliance - 13 test cases, semua passed el-pablos <yeteprem.end23juni@gmail.com>

- Test createHandlerLogger returns valid logger - Test LogEmitter.log adds entry - Test LogEmitter maintains buffer limit (1000 max) - Test LogEmitter emits events to listeners - Test logger has all methods (info, warn, error, debug, success, box) - Test listener subscribe/unsubscribe functionality - Test error handling for listener errors - Split into 3 files untuk compliance dengan max-lines-per-function el-pablos <yeteprem.end23juni@gmail.com>

el-pablos <yeteprem.end23juni@gmail.com>

…ed logging

el-pablos <yeteprem.end23juni@gmail.com>

- Buat integration tests yang comprehensive untuk stream translation - Test mencakup: 1. Stream state initialization 2. Base chunk creation dengan berbagai overrides 3. Event type helpers (findEvent & filterEvents) 4. THINKING_TEXT constant validation 5. Error translation event 6. Chunk structure validation (thinking, reasoning_opaque, tool calls) 7. State machine transitions 8. Event structure validation (semua event types) 9. Finish reason mapping NOTE: Tests ini dirancang untuk kompatibel dengan implementasi stream-translation.ts yang ada sekarang. Tests yang memerlukan handleReasoningOpaqueSignature dan handleToolCallsDelta tidak disertakan karena fungsi-fungsi tersebut belum diimplementasikan. Semua 21 tests pass dengan 74 expect() calls. el-pablos <yeteprem.end23juni@gmail.com>

… override 2M anti compact, max output 128K, fix port config el-pablos <yeteprem.end23juni@gmail.com>

…-robin pool rotation sekarang work dengan benar via messages api el-pablos <yeteprem.end23juni@gmail.com>

…ang belum ada suffix level el-pablos <yeteprem.end23juni@gmail.com>

Copilot

Pull request overview

This PR expands the server’s Copilot proxy to better support “thinking/reasoning” outputs and the native /v1/messages flow, alongside configuration and operational updates.

Changes:

Add a new Copilot “Messages API” client and route /v1/messages requests through either Messages API or Chat Completions based on model endpoint support/capabilities.
Enhance streaming translation to emit Anthropic thinking blocks from reasoning_text / reasoning_opaque.
Introduce new config knobs (default max output tokens, context override / truncation disable) and add file-based handler logging + account-pool round-robin adjustments.

Reviewed changes

Copilot reviewed 27 out of 31 changed files in this pull request and generated 15 comments.

Show a summary per file

File	Description
tsconfig.json	Excludes test files from TS compilation/typecheck scope.
src/start.ts	Uses config-derived port when it differs from CLI default.
src/services/copilot/create-messages.ts	New Copilot client for `/v1/messages` with beta header handling and streaming support.
src/services/copilot/chat-completion-types.ts	Adds `reasoning_text` / `reasoning_opaque` fields to chunk/response delta types.
src/routes/messages/stream-translation.ts	Translates reasoning fields into Anthropic “thinking” SSE content blocks.
src/routes/messages/request-payload.ts	Default `max_tokens` now comes from config (`defaultMaxOutputTokens`).
src/routes/messages/handler.ts	Major refactor: route to Messages API vs Chat Completions, quota optimizations, thinking block filtering, stream handling.
src/routes/messages/tests/thinking-mechanism.test.ts	New unit tests for thinking translation behavior.
src/routes/chat-completions/truncate-messages.ts	Adds truncation disable + context-window override; adjusts safety multiplier for Claude.
src/lib/request-context.ts	Adds helpers (`generateTraceId`, `runWithContext`, `getTraceId`) and deprecates old accessor.
src/lib/paths.ts	Expands `PATHS` constants; adds `ensureDir`.
src/lib/logger.ts	Adds file-based handler logging + buffering/retention + keeps LogEmitter wrapper.
src/lib/config.ts	Adds new config fields + getters (default output tokens, context override, disable truncation).
src/lib/account-pool.ts / src/lib/account-pool-selection.ts	Round-robin index/rotation behavior changes + logging on selection.
src/lib/tests/*	New tests for paths/logger behaviors.
src/tests/integration/stream-translation.integration.test.ts	New integration test scaffold for stream translation (currently contains outdated notes).
package.json / README.md	Large metadata/docs updates beyond “thinking” feature scope.
eslint.config.js	Ignores all tests from linting.
.github/workflows/*	CI trigger changes + new version-bump workflow.
.DS_Store / screenshots/.DS_Store	OS-generated files added (shouldn’t be committed).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-23T23:47:06Z

src/lib/logger.ts

+// Set up periodic buffer flushing
+setInterval(flushAllBuffers, FLUSH_INTERVAL_MS)
+


The module sets up a global setInterval(flushAllBuffers, ...) at import time. In Node/Bun this can keep the event loop alive and make bun test hang or become flaky. Consider creating the interval only when file logging is enabled/used, and call unref() on the timer (or provide an explicit start/stop lifecycle hook) so importing the logger doesn’t prevent process exit.

Copilot · 2026-03-23T23:47:07Z

src/routes/messages/handler.ts

+  // Apply quota optimization
  const quotaContext = applyQuotaOptimization(anthropicPayload, c)
-
  const accountInfo = getAccountInfo()
-  applyFallbackIfNeeded(anthropicPayload)
-  logRequestStart(anthropicPayload, accountInfo)

-  const translatedPayload = translateToOpenAI(anthropicPayload)
-
-  // Apply truncation to fit within model's prompt token limit
-  const openAIPayload = await truncateMessages(translatedPayload)
-
-  consola.debug(
-    "Translated OpenAI request payload:",
-    JSON.stringify(openAIPayload),
-  )
-
-  tokenState.input = estimateInputTokens(openAIPayload.messages)
+  // Apply fallback if needed
+  applyFallbackIfNeeded(anthropicPayload)



The handler still writes to requestCache in the non-streaming response path, but this flow no longer checks requestCache.get(...) before calling upstream (the earlier cache-hit handling was removed). That makes caching effectively write-only and eliminates cache hits. Consider reintroducing a cache lookup for non-streaming requests (before executing the upstream call) or removing the cache-write if caching is intentionally disabled for this route.

Copilot · 2026-03-23T23:47:07Z

eslint.config.js

+      "**/__tests__/**",
+      "**/*.test.ts",


ESLint is configured to ignore all __tests__ and *.test.ts files, so the newly added tests won’t be linted in CI. If the intent is to rely on linting for code quality, consider removing these ignores (or limiting them to generated fixtures) so tests follow the same standards as production code.

Suggested change

"**/__tests__/**",

"**/*.test.ts",

Copilot · 2026-03-23T23:47:07Z

src/__tests__/integration/stream-translation.integration.test.ts

+ * NOTE: Some tests are marked as skipped (.skip) because they depend on
+ * functions that are not yet implemented in stream-translation.ts:
+ * - handleReasoningOpaqueSignature
+ * - handleToolCallsDelta
+ *
+ * These tests will be enabled once those functions are implemented.


The header comment says translateChunkToAnthropicEvents references missing functions (handleReasoningOpaqueSignature, handleToolCallsDelta) and that tests are skipped until they’re implemented. In the current code, handleReasoningOpaqueSignature exists and tool calls are handled via handleToolCallsLoop, so this note is now misleading. Please update/remove the comment (and consider adding a real integration test that calls translateChunkToAnthropicEvents end-to-end).

Suggested change

* NOTE: Some tests are marked as skipped (.skip) because they depend on

* functions that are not yet implemented in stream-translation.ts:

* - handleReasoningOpaqueSignature

* - handleToolCallsDelta

*

* These tests will be enabled once those functions are implemented.

* These tests aim to reflect the current implementation of the stream

* translation logic, including reasoning-opaque handling and tool calls.

Copilot · 2026-03-23T23:47:08Z

.github/workflows/ci.yml

 on:
  push:
-    branches: ['*']
+    branches: [main, develop]
  pull_request:
-    branches: ['*']
+    branches: [main, develop]


CI is now configured to run on pushes only to main and develop. This removes CI coverage for pushes to other branches (feature branches, release branches, etc.), which can allow breakages to go unnoticed until a PR is opened/retargeted. If that’s not intentional, consider restoring push triggers for all branches (or at least ** with excludes).

Copilot · 2026-03-23T23:47:09Z

src/lib/__tests__/logger-emitter.test.ts

+/**
+ * Unit tests for LogEmitter
+ */
+
+import { describe, expect, it, mock } from "bun:test";
+
+// Mock the state module
+void mock.module("~/lib/state", () => ({
+  state: {
+    verbose: false,
+  },
+}));
+


This test file uses semicolons and a formatting style that differs from the rest of the repo’s tests. Since tests are currently excluded from ESLint/typecheck, these inconsistencies (and potential TS issues) won’t be caught in CI. Consider reformatting to match the repo style and re-enabling linting/typechecking for test files.

Copilot · 2026-03-23T23:47:09Z

src/lib/logger.ts

+const LOG_RETENTION_DAYS = 7
+const LOG_RETENTION_MS = LOG_RETENTION_DAYS * 24 * 60 * 60 * 1000
+const CLEANUP_INTERVAL_MS = 24 * 60 * 60 * 1000
+const LOG_DIR = path.join(PATHS.APP_DIR, "logs")


LOG_DIR is derived from PATHS.APP_DIR (~/.local/share/.../logs), but PATHS also defines LOGS_DIR under ~/.config/.../logs and the rest of the code/docs appear to treat ~/.config as the config/log area. Consider using PATHS.LOGS_DIR (and ensuring it exists) to avoid scattering storage locations and duplicating path logic.

Suggested change

const LOG_DIR = path.join(PATHS.APP_DIR, "logs")

const LOG_DIR = PATHS.LOGS_DIR

Copilot · 2026-03-23T23:47:09Z

src/routes/messages/handler.ts

+function generateRequestIdFromPayload(
+  payload: AnthropicMessagesPayload,
+  sessionId?: string,
+): string {
+  const content = JSON.stringify(payload.messages.slice(-3))
+  const input = (sessionId ?? "") + content + Date.now().toString()
+  return Buffer.from(input).toString("base64").slice(0, 32)
+}


generateRequestIdFromPayload is described as being for deduplication, but it includes Date.now(), making the output non-deterministic even for identical payloads. If this ID is meant to dedupe/replay-protect requests across retries, generate it deterministically from stable inputs (e.g., sessionId + message content hash) or rename it to reflect that it’s just a unique-per-request ID.

Copilot · 2026-03-23T23:47:09Z

src/lib/account-pool-selection.ts

-  return activeAccounts[index]
+  const selected = activeAccounts[index]
+  poolState.currentIndex++
+  consola.info(


selectRoundRobinAccount logs the selected account at info level on every request. This can generate a lot of log noise under load and may leak account identifiers into logs unnecessarily. Consider switching this to debug (or gating behind state.verbose) and/or sampling.

Suggested change

consola.info(

consola.debug(

Copilot · 2026-03-23T23:47:10Z

package.json

+  "description": "GitHub Copilot API Proxy dengan fitur enterprise-grade untuk integrasi Claude Code. Mendukung thinking mechanism, multi-account pool, request caching, dan file-based logging.",
  "keywords": [
-    "proxy",
    "github-copilot",
-    "openai-compatible"
+    "claude-code",
+    "api-proxy",
+    "anthropic",
+    "openai",
+    "thinking-mechanism",
+    "enterprise",
+    "typescript"
  ],
-  "homepage": "https://github.com/prassaaa/copilot-api",
-  "bugs": "https://github.com/prassaaa/copilot-api/issues",
+  "homepage": "https://github.com/el-pablos/copilot-api#readme",
+  "bugs": {
+    "url": "https://github.com/el-pablos/copilot-api/issues"
+  },
  "repository": {
    "type": "git",
-    "url": "git+https://github.com/prassaaa/copilot-api.git"
+    "url": "https://github.com/el-pablos/copilot-api.git"
  },
-  "author": "Prasetyo Ari Wibowo",
+  "license": "MIT",
+  "author": "el-pablos <yeteprem.end23juni@gmail.com>",


This PR changes package metadata (description/homepage/repository/author) to different owners/URLs, which doesn’t match the PR description about adding thinking/model support. If this isn’t an intentional rebranding/fork handover, it should be reverted or moved to a separate PR to avoid surprising downstream consumers (npm metadata, links, attribution).

@copilot open a new pull request to apply changes based on this feedback

…k, model types + dokumentasi round-robin bug fix el-pablos <yeteprem.end23juni@gmail.com>

el-pablos added 28 commits March 21, 2026 09:05

config: update gitignore untuk proteksi credentials dan secrets

f6214b3

el-pablos <yeteprem.end23juni@gmail.com>

add: nambahin field reasoning_text dan reasoning_opaque ke interface …

2f2e5a0

…Delta dan ResponseMessage el-pablos <yeteprem.end23juni@gmail.com>

config: update package.json dengan metadata dan deskripsi lengkap

ee6fb9b

el-pablos <yeteprem.end23juni@gmail.com>

update: enhance logger dengan file-based logging dan auto cleanup

18ac386

el-pablos <yeteprem.end23juni@gmail.com>

update: tambahin safety multiplier khusus buat claude models

9a76fb5

el-pablos <yeteprem.end23juni@gmail.com>

test: tambahin test case buat claude safety multiplier

34c9b81

el-pablos <yeteprem.end23juni@gmail.com>

docs: nambahin dokumen analisis perbandingan cina-copilot vs copilot-api

8a56440

el-pablos <yeteprem.end23juni@gmail.com>

docs: nambahin mega prompt implementasi thinking mechanism

9a573de

el-pablos <yeteprem.end23juni@gmail.com>

test: nambahin unit test untuk request context module

022b67f

el-pablos <yeteprem.end23juni@gmail.com>

add: nambahin path baru untuk logs, cache, dan config dir

e4698fc

el-pablos <yeteprem.end23juni@gmail.com>

docs: nambahin dokumentasi caching analysis

b7c4dee

el-pablos <yeteprem.end23juni@gmail.com>

docs: update README dengan struktur yang lebih lengkap

cc14412

el-pablos <yeteprem.end23juni@gmail.com>

config: exclude test files dari typescript compilation

30037bd

el-pablos <yeteprem.end23juni@gmail.com>

test: nambahin integration test buat stream translation

5abb085

el-pablos <yeteprem.end23juni@gmail.com>

config: hapus test file yang dipindah lokasi

5ede90f

el-pablos <yeteprem.end23juni@gmail.com>

update: refactor stream translation dengan thinking block handler

da9511d

el-pablos <yeteprem.end23juni@gmail.com>

config: format ulang logger-emitter test pake prettier

3832a2e

el-pablos <yeteprem.end23juni@gmail.com>

test: update logger-emitter test formatting

756d052

el-pablos <yeteprem.end23juni@gmail.com>

test: update logger emitter test untuk kompatibilitas dengan file-bas…

262573e

…ed logging

config: exclude test files dari eslint checking

55bff6b

el-pablos <yeteprem.end23juni@gmail.com>

config: format integration test pake prettier

56d1163

el-pablos <yeteprem.end23juni@gmail.com>

test: nambahin unit test untuk thinking mechanism di stream translation

72e2264

feat: tambahin auto-inject extended thinking buat claude 4.5, context…

3a07296

… override 2M anti compact, max output 128K, fix port config el-pablos <yeteprem.end23juni@gmail.com>

fix: perbaiki token tracking 0/0 di request history streaming + round…

e16750e

…-robin pool rotation sekarang work dengan benar via messages api el-pablos <yeteprem.end23juni@gmail.com>

Copilot AI review requested due to automatic review settings March 23, 2026 23:42

Copilot started reviewing on behalf of el-pablos March 23, 2026 23:42 View session

feat: tambahin auto-inject reasoning_effort xhigh untuk model gpt-5 y…

c7fae82

…ang belum ada suffix level el-pablos <yeteprem.end23juni@gmail.com>

Copilot AI reviewed Mar 23, 2026

View reviewed changes

chore: update webui dashboard, types, translation layer, version chec…

8a8d3a6

…k, model types + dokumentasi round-robin bug fix el-pablos <yeteprem.end23juni@gmail.com>

el-pablos closed this Mar 24, 2026

el-pablos mentioned this pull request Mar 24, 2026

Dev tama #7

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Dev tama#6

Dev tama#6
el-pablos wants to merge 30 commits intoprassaaa:mainfrom
el-pablos:dev-tama

el-pablos commented Mar 23, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

Copilot AI Mar 23, 2026

Uh oh!

el-pablos Mar 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		// Set up periodic buffer flushing
		setInterval(flushAllBuffers, FLUSH_INTERVAL_MS)

	const LOG_DIR = path.join(PATHS.APP_DIR, "logs")
	const LOG_DIR = PATHS.LOGS_DIR

Conversation

el-pablos commented Mar 23, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

el-pablos Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants