feat: Autonomous GitHub Agent (ReAct framework)#1
Merged
Conversation
- Created Pydantic settings with type-safe configuration - Implemented structured logging with structlog - Built cost tracking system for API usage - Added rate limiter with token bucket algorithm - Implemented retry utilities with exponential backoff - Created state manager with SQLite operations - Built base classes for researchers and publishers - Added database schema with tables for: - published_items (content tracking) - newsletters (publication history) - publishing_logs (per-platform status) - api_metrics (cost tracking) - content_fingerprints (deduplication) All Agent 1 (Data Layer) tasks complete. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Created ArXiv researcher extending BaseResearcher - Fetches from ArXiv RSS feed with relevance scoring - Implements time window filtering (last hour) - Scores based on keywords: LLMs, multimodal, agents, etc. - Created research-arxiv skill with detailed workflow - Created content-researcher subagent specification - Handles deduplication and error recovery Research layer ready for testing. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Created src/main.py with test and production modes - Built scripts/test_foundation.py to verify components - Updated STATUS.md with current progress - Added __init__.py files for proper package structure Foundation ready for testing and MCP server development. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- IMPLEMENTATION_PROGRESS.md: Detailed day 1 summary - PROJECT_STRUCTURE.md: Complete project structure overview - Includes statistics, next steps, and navigation guide Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Fixed Settings to allow optional ANTHROPIC_API_KEY for testing - Created pytest.ini with test configuration and markers - Built tests/conftest.py with shared fixtures - Implemented tests/unit/test_state_manager.py (complete unit tests) - Created test-planner agent for comprehensive test case planning - Added TESTING_GUIDE.md with best practices and workflows - Fixed test_foundation.py to work without real API keys - Added production config validation Testing workflow now supports: - TDD (test-driven development) - TAD (test-after development) - Hybrid approach - Test planning agent integration Run: pytest -v (for unit tests) Run: python scripts/test_foundation.py (for quick check) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Changed ArXiv RSS URL from HTTP to HTTPS (fixes 301 redirect) - Fixed syntax error in test_state_manager.py (async import) - Updated CLAUDE.md with critical venv activation instructions - Added venv reminder to all command examples All tests now passing: - Foundation test suite: ✓ - Pytest unit tests: 9/9 passed ✓ Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Created Database MCP Server with 3 tools: - check_duplicate: Check if content exists - store_content: Save published items - get_metrics: Retrieve API usage stats - Uses official MCP Python SDK - Wraps StateManager for all database operations - Implements proper error handling and logging - Full async support Tests: 7/7 passed (0.70s) - Server initialization - Duplicate checking (exists/not exists) - Content storage (success/validation errors) - Metrics retrieval (empty/with data) This unblocks Claude integration with database! Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Documentation updates: - STATUS.md: Marked MCP server complete, updated metrics (55% complete) - PROJECT_STRUCTURE.md: Marked database_server.py as done - Updated test counts and file statistics Testing improvements: - Created scripts/test_mcp_server.py for end-to-end MCP testing - Tests all 3 MCP tools (check_duplicate, store_content, get_metrics) - Tests error handling and validation - All tests passing ✓ Test Results: - Pytest: 16/16 tests passing - Foundation test: All passing - MCP server test: All passing Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Documentation restructure:
- Created docs/ folder for all project documentation
- Moved STATUS.md, IMPLEMENTATION_PROGRESS.md, PROJECT_STRUCTURE.md, TESTING_GUIDE.md to docs/
- Added docs/README.md as documentation index
- Updated CLAUDE.md with documentation references
- Updated root README.md with new doc locations
Benefits:
✓ Cleaner root directory
✓ All documentation in one organized location
✓ Clear documentation index for easy navigation
✓ CLAUDE.md automatically references docs
File structure:
/
├── README.md # Project overview
├── .claude/CLAUDE.md # Dev guidelines (auto-read)
└── docs/ # All project documentation
├── README.md # Documentation index
├── STATUS.md # Current progress
├── PROJECT_STRUCTURE.md # File organization
├── TESTING_GUIDE.md # Testing practices
└── IMPLEMENTATION_PROGRESS.md # Detailed progress
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Added quick-start section for developers resuming work in fresh sessions: - Simple command to get oriented - Current phase summary - Pre-coding checklist - Direct link to next steps Makes it easy to pick up where we left off without reading entire file. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Created 4 skills: session-start, session-end, log-session, update-status - Implemented session handover log system in docs/logs/ - Compressed STATUS.md from 345 to 94 lines (73% reduction) - Added agent selection rubric to CLAUDE.md for autonomous mode selection - Archived historical session docs to logs/ with new format - Updated README.md and START_HERE.md with /session-start workflow Enables perfect session continuity with one-command automation. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Implemented ContentEnhancer to orchestrate 4 AI enhancement agents: - HeadlineWriter: Generates viral headlines (Claude Sonnet) - TakeawayGenerator: Creates 'why it matters' insights (Claude Haiku) - EngagementEnricher: Extracts social proof metrics (local) - SocialFormatter: Formats category messages (Claude Haiku) Features: - Sequential item enhancement with exponential backoff retry (3 attempts) - Template fallback when AI enhancement fails (never fails) - Category grouping (max 5 items per category, sorted by relevance) - AI-powered category formatting with simple fallback - Comprehensive cost and performance metrics tracking Integration: - Added TelegramPublisher.publish_enhanced() for enhanced publishing - Added TelegramFormatter.format_enhanced() for enhanced formatting - Backward compatible: Original publish_newsletter() unchanged Testing: - 10 ContentEnhancer unit tests (sequential flow, retry logic, fallbacks) - 6 TelegramFormatter enhanced tests (formatting, splitting, markdown) - 5 integration tests (end-to-end enhancement + publishing) - All 21 tests passing Cost: ~$0.035 per 15-item newsletter (well under $3/day target) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Fixed issue where ANTHROPIC_API_KEY wasn't loaded when running main.py from src/ directory. Pydantic's env_file was using a relative path '.env' which looked in the current working directory instead of project root. Solution: Compute project root before model_config and use absolute path for env_file. Now works from any directory. Fixes: 'ANTHROPIC_API_KEY is required for production' error Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Use explicit AsyncAnthropic import for consistency with ContentPipeline - Changes in headline_writer.py, takeaway_generator.py, social_formatter.py - From: import anthropic; anthropic.AsyncAnthropic() - To: from anthropic import AsyncAnthropic; AsyncAnthropic() - Improves code clarity and IDE support - All 10 ContentEnhancer unit tests passing Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Add enhance_phase() between filter and publish phases: Research → Filter → Enhance (optional) → Publish → Record Features: - Feature flag: settings.enable_content_enhancement (default: True) - Feature flag: settings.max_items_per_category (default: 5) - Backward compatible: publishers without publish_enhanced() supported - Metrics tracking: cost, success rate, AI vs template - Cost control: orchestrator checks budget before enhancement Architecture: - ContentEnhancer is centralized, reusable across platforms - TelegramPublisher.publish_enhanced() called when available - Fallback to publish_newsletter() for other publishers - Enhancement metrics tracked in CycleResult Implementation: - Add enhance_phase() method to Orchestrator - Update publish_phase() to handle both Newsletter and CategoryMessage - Add _category_to_newsletter() helper for backward compatibility - Update record_phase() to log enhancement metrics - Add enhancement fields to CycleResult dataclass Testing: - scripts/test_content_enhancer_real.py - Test with real sources * Fetches from ArXiv, HuggingFace, Reddit, TechCrunch * 100% AI enhancement success rate * Cost: $0.05 per 20 items - scripts/test_orchestrator_enhanced.py - Full orchestrator cycle * Validates enhancement integration * Tracks metrics correctly Cost Impact: - +$0.035/newsletter (15 items, 5 categories) - Budget: $0.84/day (24 cycles) = 28% of $3 daily budget Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Session Documentation: - Created session log for 2026-02-18-1 (orchestrator integration) - Updated STATUS.md (Phase 2B → Phase 3, 100% complete) - Updated .gitignore to track docs/logs/ (session logs now versioned) - Added historical session logs (2026-02-16, 2026-02-17) - Minor CLAUDE.md update (planning code guidance) Session Highlights: - Fixed AsyncAnthropic imports (consistency) - Tested with real sources (100% AI success, $0.05 cost) - Integrated ContentEnhancer into orchestrator pipeline - Added feature flags (enable_content_enhancement, max_items_per_category) - Created 2 integration test scripts - Made 2 commits (bug fixes + integration) Architecture: Research → Filter → Enhance (optional) → Publish → Record Next Steps: - End-to-end test with real Telegram - Monitor enhancement quality (1 day) - Adjust prompts if needed Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add pyproject.toml with ruff, mypy, and pytest config (replaces pytest.ini) - Add requirements-dev.txt with dev tooling dependencies - Add .pre-commit-config.yaml with ruff v0.15.1, formatting, and detect-private-key hooks - Add GitHub Actions CI workflow (lint + unit-tests 3.10/3.11 + integration + gitleaks secret scan) - Add GitHub Actions auto-PR workflow for agent-* branches with auto-merge - Add 62 new unit tests: test_researchers.py (41), test_utils.py (21) - Fix 2 pre-existing test failures in test_orchestrator.py (mock AsyncMock gaps) - Apply ruff format and fix 308 lint issues across all source files Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Stage 5 now automatically commits documentation without asking. Removes the optional prompt; commit runs unconditionally with Co-Authored-By trailer. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- image_generator.py: Use Path | None for optional arg (PEP 604 syntax) - markdown_formatter.py: Add type annotation for by_category dict - telegram_formatter.py: Add type annotation for current list Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- state_manager.py: annotate metrics dict as dict[str, Any] - content_pipeline.py: assert client not None before API call - markdown_publisher.py: fix publish/publish_newsletter signatures to be compatible with BasePublisher; add dict-to-Newsletter conversion - orchestrator.py: add type annotations, assert enhancer not None, use isinstance narrowing for gather results, handle None published_date Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implements a fully autonomous GitHub PR lifecycle agent:
- AgentLoop ABC (src/agents/base.py): ReAct loop with poll→triage→act→record,
run_forever() with max_cycles support for testing
- GitHubClient (src/github/client.py): httpx REST wrapper for GitHub API
(9 methods, rate-limited via token bucket)
- GitHubMonitor (src/github/monitor.py): orchestrates the full PR lifecycle
- 3 AI workers (src/github/workers/):
- PRDescriber: Claude Haiku generates PR descriptions
- CIFixer: 3-tier (ruff auto-fix → Claude Sonnet → secret alert)
with circuit breaker at MAX_FIX_ATTEMPTS=3
- CodeReviewer: Claude Sonnet posts reviews on green CI, idempotent
via <!-- elvagent-review --> marker
- github_events DB table + 3 StateManager methods for deduplication
- --cycles N flag for controlled test runs
Verified live against GitHub API: token auth, rate limiting, PR polling
all working. 217 tests passing.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add investigation layer: CI logs + check annotations + local file contents - Escalate Tier 1 (ruff) to Claude Sonnet when ruff makes no changes - Add get_fix_history() to StateManager for per-PR fix context - Refactor CIFixer tests to use _make_fixer() factory pattern - Add test for ruff→Claude escalation path Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
run_github_monitor() used `logger` before main() set it up as a global.
Add get_logger("main") at module level to satisfy mypy.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
What's included
New modules
src/agents/base.pyAgentLoopABC:poll → triage → act → recordloop +run_forever(max_cycles=N)src/github/client.pyGitHubClient: httpx REST wrapper (9 methods, token-bucket rate-limited)src/github/monitor.pyGitHubMonitor(AgentLoop): orchestrates the PR lifecyclesrc/github/workers/pr_describer.pysrc/github/workers/ci_fixer.pysrc/github/workers/code_reviewer.pyModified files
src/config/constants.py— GitHub rate limit + agent constantssrc/config/settings.py— 6 new GitHub settings (github_token,github_repo, etc.)src/core/state_manager.py—github_eventstable +record_github_event,is_github_event_processed,count_fix_attemptssrc/main.py—--mode=github-monitor --cycles=Nentry pointTest plan
ruff check+ruff formatcleanmypy src/ --ignore-missing-importsCI-greenpython src/main.py --mode=github-monitor --verbose --cycles=1Usage
🤖 Generated with Claude Code