agenthropic

disler/claude-code-hooks-multi-agent-observability — C−

Panel: B+ (v1) / 2.7 (v2) · Independent: C− ▼▼ · Role: teaching example only.

The most-starred candidate (1,475★ / 385 forks, vs the report’s “~1,400 / 372”), and the biggest gap between reputation and substance. Stalled 2026-02-08.

The subagent data is a dead path

• The hook sends agent_id / agent_type, but the server drops them (db.ts:127) — they survive only inside an opaque JSON blob, not as columns.
• Schema is a single events table: no parent column, and no graph library anywhere (grep finds no d3 / cytoscape / dagre / reactflow).
• “Trace every task handoff across the swarm” is, in reality, flat swim-lanes keyed on an application label. No hierarchy is reconstructable from what it stores.

Quality & legal blockers

• Zero tests.
• No license — private:true, no license field → all-rights-reserved. A hard OPCⁿ blocker.
• Effectively one real runtime dependency (Vue) plus two dead deps.
• Stalled since 8 Feb 2026.

Security — unfit for exposure

• Unauthenticated POST /events; CORS *.
• SSRF: the server dials an arbitrary responseWebSocketUrl taken from the request payload (index.ts:198-201).
• The .env / key guard the report credited is commented out (pre_tool_use.py:324-327).

The one thing it’s good for

• send_event.py (~180 lines) is the clearest teaching example of the whole ingest pattern: hook → HTTP → SQLite → WebSocket → browser. Read it to learn the loop; do not build on it.

Verdict

Popular, but hollow: no hierarchy, no tests, no license, stalled, and insecure. Learn the pattern from its hook script and move on. C−.

This site is open source. Improve this page.