English · 中文

Alert handling, split into three services that each do one job — and an agent runner (hookprobe) that is useful on its own, whether or not you care about alerts.

MIT licensed. docker compose up. → github.com/itswl/hookstack

hookprobe: an agent run behind an HTTP contract

You POST a task. hookprobe runs a single tool-using agent session (Claude Agent SDK: bash, MCP servers, web search, SKILL.md skills) and serves the report to whoever polls for it. No channels, no device pairing, no chat history.

It speaks an OpenClaw-compatible dialect, so a client that already treats that gateway as its analysis backend switches by changing a URL.

git clone https://github.com/itswl/hookstack && cd hookstack
printf 'HOOKPROBE_TOKEN=change-me\nANTHROPIC_API_KEY=sk-ant-...\n' > .env
docker compose --env-file .env -f hookprobe/deploy/docker-compose.yml up -d --build

curl -s -X POST localhost:8088/hooks/agent \
  -H "Authorization: Bearer change-me" -H 'Content-Type: application/json' \
  -d '{"message":"Which processes are listening, and on what?","sessionKey":"demo:1"}'

curl -s -H "Authorization: Bearer change-me" localhost:8088/sessions/demo:1/final

Three things that make it different

It terminates. /final answers 202, or a 200 that is final — including when the run crashed or timed out, where the report’s root_cause names the runner failure. A poller writes the result on its first confirming read and never invents stability heuristics.

The agent cannot edit what steers the next run. .claude/ (skills, roles, settings), CLAUDE.md and the audit log are closed to it — by a PreToolUse hook that refuses the write, and by a digest of every input file compared before and after each run, because the two fail differently. Without this, one injected line reaching .claude/skills/ outlives the run that read it and comes back as the operator’s own runbook.

Finished runs leave runbooks behind. A completed investigation distills its own record into a SKILL.md, written by the service and never through the agent’s tools. The second investigation of the same condition adds a case rather than replacing what was there, and every write — by a run or by a person — snapshots what it displaced first.

The sessions console

Every tool call of an investigation, as it happens

The diagnostic runbook a run distilled for itself

The whole family

Component Role Deliberately does NOT
hookrelay the pipe — adapts every upstream dialect in and every channel format out, and accounts for all of it understand content, or judge
hookjudge the judge — one event in, one verdict out, four routes ordered by cost render cards, or know channels
hookprobe the investigator — one read-only agent run per alert that earns it receive alerts, or send notifications 
upstream alert sources (Grafana / Alertmanager / cloud monitoring …)
      │
      ▼
  hookrelay :8100 ──► hookjudge :8200 ──► hookrelay ──► Feishu / DingTalk / WeCom
  (pipe: adapt+route+ledger) │ (judge: verdict+cost)   (formats and delivers)
                             │
                             └──► hookprobe :8088 ──► hookrelay ──► the same channels
                                  (investigator: read-only)  /hook/probe-notify

hookrelay's ledger: every message accounted for, every delivery with an outcome

hookjudge's status page: four verdicts with their routes

Every screenshot above comes from one local Docker run started from nothing — not mockups.

Read more