agenthropic

simple10/agents-observe — A−

Panel: B+ (v1) / 4.1, highest (v2) · Independent: A− ▲ · Role: recommended base.

Metrics (2026-07-03): 607★ / 58 forks, pushed 29 Jun, MIT + real LICENSE (© 2025 Joe Johnston). The report’s “rate-limited / stale 5 Jun / no true DAG” read was wrong on all three counts.

Why it wins

It has the subagent graph the report said it lacked

• constellation/agent-tree.ts → buildAgentTree() builds a real parent→child subagent tree: orphan-reparenting, root synthesis, drawn edges.
• A bespoke dependency-free N-body live force-directed graph renders it.
• Edges are derived from the event stream, session-scoped (not persisted as first-class rows) — the one real extension point for a global/historical DAG.

This single fact refutes the panel’s tie-break (see ../report-meta-audit.md §1).

Tests are real

• 78 test files, 1,985 expect() calls. agent-tree.test.ts asserts nesting and reparenting behaviour directly.
• Caveats: no test-on-PR CI (tests run only via just check), and the report’s “semantic-release CI” claim is false — releases are a hand-rolled release.sh.

Docker is avoidable (the decisive operational trade-off)

• The plugin happy-path is hard-Docker, but AGENTS_OBSERVE_RUNTIME=local / just start-local runs pure Node + native better-sqlite3, no daemon.
• Fits the Mac Mini constraint (no always-on Docker next to Ollama/Telegram) → run under launchd.

Clean, forkable architecture

• Ports/adapters storage; strategy-pattern agent classes (Claude Code + Codex + a hermes variant) → extensible to new agent runtimes.
• 5-table schema: projects, sessions, agents, events, filters. Hierarchy is derived from the event stream rather than stored — clean, but persist it if a global DAG is wanted.
• arm64 multi-arch image; runs natively on M4.
• Highest overall forkability for an OPCⁿ product of the whole set.

Must-fix before exposure

• Binds 0.0.0.0, wildcard CORS, no auth — LAN-exposed as shipped.
• Stores full tool payloads (redact).
• First patch: bind 127.0.0.1, add a DASHBOARD_TOKEN, same-origin WS check. See ../security.md.

What’s missing (build backlog)

• No Telegram sink → graft hoangsonww’s formatTelegram.
• No dollar-cost / delegation-savings → graft cast’s metric.
• Session-scoped, non-persisted edges → persist + add a global historical DAG view.
• Single-host → cross-machine aggregation is future work.

Verdict

The disciplined option the panel under-rated: real graph, real tests, clean license, Docker-optional, most forkable. A−, held back from A only by the 0.0.0.0/no-auth posture and the non-persisted, session-scoped edges. Recommended base — see ../recommendation.md.

This site is open source. Improve this page.