asepharyana
99d47c8ff8
Claude Code could not complete AI fixes / conflict resolution against this host's provider setup: it hung 900s spawning an MCP server, then failed with 'body is JSON but not a Message' (Anthropic-Messages transport mismatch), then exited 1 with empty stderr. The Hermes gateway already runs continuously with the working 9router provider config and a full toolset, so drive it directly: - POST http://127.0.0.1:8642/v1/chat/completions (OpenAI-compatible API server) - bearer auth from API_SERVER_KEY (env or ~/.hermes/.env), overridable via API_SERVER_URL - model_options.max_turns caps a runaway run; 900s timeout for sync, 600s for PR fixes - every failure maps to an [INFRA] string so the existing skip-once logic works - the agent commits locally; the WORKER pushes (agents must never push) run_ai_fix no longer shells out to claude; it calls the API server, then pushes the agent's commit itself and reports push failures explicitly. Sync call sites keep their contract via _run_claude_sync -> _run_hermes_sync alias, with labels renamed hermes_sync_conflicts / hermes_sync_quality. Tests: 59/59 (8 new assertions exercise a real local HTTP round-trip: path, bearer auth, OpenAI message shape, max_turns cap, HTTP-error/missing-key/ unreachable -> [INFRA]).
PR-Agent Server
Nix-deployed GitHub App server for automated PR review + auto-merge using a custom LLM endpoint (9router/Omniroute).
Architecture
GitHub webhook → Caddy (reverse proxy, :4002)
→ pr-agent-server (Nix profile, uvicorn on :4002)
→ PR-Agent github_app.py (FastAPI)
→ 9router API (custom OpenAI-compatible endpoint)
Project Layout
pr-agent-server/
├── src/ # Main application modules
│ ├── run_server.py # FastAPI server + analytics/metrics + Discord webhook
│ ├── auto_merge_bot.py # Periodic PR review→approve→merge bot
│ ├── trivial_merge.py # Trivial PR fast-path (docs/dependabot/tiny diffs)
│ ├── health-check.py # Model health watchdog (tests primary + fallbacks)
│ ├── sync-key.py # Auto-syncs BWS router key to disk on service start
│ ├── callback_server.py # Dev callback server for GitHub App manifest
│ ├── start_server.py # Legacy server start script
│ └── config/ # Runtime config (gitignored at deploy time)
├── scripts/ # Setup and deployment helpers
│ ├── setup_all.py # Full setup: manifest + config + systemd service
│ ├── setup_app.py # App-specific setup
│ ├── generate_manifest.py # GitHub App manifest URL generator
│ └── generate_manifest_domain.py
├── templates/
│ └── manifest.json # GitHub App manifest template
├── .github/workflows/
│ ├── deploy.yml # CI: syntax → build → deploy → GC
│ ├── flakehub-publish-rolling.yaml
│ └── mirror-gitea.yml
├── flake.nix # Nix build (creates venv + binary wrappers)
├── flake.lock # Pinned Nix dependencies
├── .editorconfig # Editor formatting rules
├── .gitignore
└── README.md
Development
Prerequisites
- Bun 1.3.14+ (runtime + build)
- GitHub App credentials (App ID, private key, webhook secret)
- 9router/OpenAI-compatible key for the LLM
Local testing
cd server
bun install
bunx tsc --noEmit # typecheck
bun test # unit tests (16 tests)
bun src/cli.ts --tool review --repo <owner>/<repo> --pr <n> --no-publish
Run server (after setting up secrets)
# Secrets are resolved at startup: PR_AGENT_APP_ID, private key path,
# omniroute key file (see src/config.ts + src/secrets.ts)
cd server
bun src/index.ts # starts on PORT (default 4023)
Test tools end-to-end (real GitHub + LLM)
bun e2e.ts --repo asepharyana/nextjs-template --pr 19 --publish
bun src/cli.ts --tool describe --repo <owner>/<repo> --pr <n>
bun src/cli.ts --tool improve --repo <owner>/<repo> --pr <n>
Deployment
Deploy is fully automated via GitHub Actions on push to main:
# .github/workflows/deploy.yml
1. build-and-deploy → bun install → typecheck → tests → bun build --compile
→ scp binary to VPS → swap /opt/pr-agent-server/bin/pr-agent-bun
→ restart pr-agent-bun.service → health check on :4023
2. cleanup → nix-gc-vps.sh (cleans legacy Nix store entries)
The production server is a single compiled binary
(/opt/pr-agent-server/bin/pr-agent-bun) running as a systemd service
(pr-agent-bun.service, port 4023, secrets via bws-exec pr-agent).
Secrets required in GitHub Actions:
VPS_HOST— VPS IP addressVPS_USER— SSH userSSH_PRIVATE_KEY— SSH private key for deploy userGITEA_TOKEN— for Gitea mirror (if using mirror workflow)
Ops
- Health watchdog: cron
pr-agent-health-watchdog(every 10 min) →~/.hermes/scripts/pr-agent-health-check.sh→ curlhttp://127.0.0.1:4023/health - Secrets: systemd
ExecStart=/usr/local/bin/bws-exec pr-agent env PORT=4023 /opt/pr-agent-server/bin/pr-agent-bun - Prometheus:
GET /api/metrics→pr_agent_requests_total,pr_agent_model_failures - Analytics:
GET /api/analytics→ JSON summary (legacypr-agent.*.log+pr-agent.bun.jsonl) - Discord:
POST /api/v1/notify_review→ pr-agent-ops webhook
Legacy (Python/Nix — retired 2026-09-21)
The original Python pr_agent server (FastAPI + Nix build, port 4002) is fully
retired: systemd unit deleted, venv removed, src/*.py + scripts/setup_* +
flake removed from the repo. The Bun binary replaced it end-to-end.
License
MIT — see LICENSE if present at deploy.