agenthropic

hoangsonww/Claude-Code-Agent-Monitor — B−

Panel: A− (v1) / 4.0 (v2) · Independent: B− ▼ · Role: study & harvest, don’t adopt blind.

The panel’s primary pick. Genuinely the most feature-complete out of the box — and the most dangerous, most oversold, and least maintainable of the serious options.

Metrics: 92,163 LOC (exact), MIT + LICENSE file. README is 2,137 lines with 70 badges — inflation confirmed.

The real, well-built core (credit due)

• 12-table schema; dual SQLite driver — better-sqlite3 with a node:sqlite fallback.
• 65 real test files: 1,478 server + 427 client assertions.
• Nested subagent tree is real: SessionDetail.tsx:676-810 recursively renders renderAgentNode, backed by an agents.parent_agent_id column. Issue #200 (nested-hierarchy fix) is the current HEAD merge.

But the “DAG cockpit” is oversold

• OrchestrationDAG.tsx is a type-aggregated, 3–4-layer diagram — categories of agents, not a per-instance orchestration graph.
• The true nesting is a collapsible indented tree, reconstructed post-hoc on SubagentStop. Impressive, but not the live per-instance DAG the report’s prose implies.

The RCE — real, and the report mis-diagnosed it

• /api/run (server/routes/run.js) accepts a permission-mode from the browser request body, and ALLOWED_PERMISSION_MODES includes bypassPermissions (run.js:96).
• Result: a browser request spawns claude --permission-mode bypassPermissions in an attacker-chosen absolute cwd → code execution as the host user.
• DASHBOARD_TOKEN auth is opt-in and a no-op when unset (security.js:133). On 0.0.0.0 without a token, this is a self-hosted RCE box.
• The report blamed a “concurrency cap of 10,000” — a red herring; the lever is the permission mode.
• Good news: the spawner is cleanly excisable — ~6 files + one mount line + one table. Delete it and the RCE is gone.

Its genuine edge for Ivan

• First-class formatTelegram provider (webhook-providers.js:177) plus alert_rules / webhook_targets / webhook_deliveries schema — the easiest Telegram bridge in the whole set. This is the piece worth harvesting into the simple10 base.

Maintainability red flag

• Bus factor = 1. All 208 source files carry @author Son Nguyen.
• 70 badges, a 207 KB ARCHITECTURE.md, a 202 KB landing page — enterprise cosplay over a solo project. Adopting 92k LOC from a single author is a real support burden.

Verdict

Feature-rich and genuinely capable, but oversold on the DAG, carrying a live RCE, and maintained by one person. B−. Keep it cloned as a reference for its D3 graph polish and harvest its Telegram provider; adopt wholesale only if out-of-box richness today outweighs clean ownership — and only after deleting the spawner. See ../recommendation.md.

This site is open source. Improve this page.