From agent-teams
Machine-wide dashboard of agent-teams initiatives. Renders every registered initiative as a compact table (status icon, id, title, phase, where) and footnotes the questions for any parked waiting on a human. Use when asked "what's running", "what needs me", "initiative status", or /initiatives.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-teams:initiativesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
**The `ateam` tool.** `ateam` is on PATH — it ships as a prebuilt binary in the plugin's `bin/` (auto-added to PATH; installed/verified by `/setup-agent-teams`). Call it as bare `ateam` everywhere.
The ateam tool. ateam is on PATH — it ships as a prebuilt binary in the plugin's bin/ (auto-added to PATH; installed/verified by /setup-agent-teams). Call it as bare ateam everywhere.
Render the initiative dashboard from the global workspace. If ateam ws fails or ateam is not found, say so and point at /setup-agent-teams.
Open initiatives:
ateam list-json
Each element includes id, title, description (the full line-oriented registry schema), labels, and notes. The description field contains the branch:, team:, and latest phase state. The labels array carries gate signals — apply the kind-resolution rule to determine gate kind:
labels contains gate:review → REVIEW (PR delivered, awaiting merge)labels contains gate:question, OR contains human but no gate:* label → QUESTION (backward-compat: pre-existing gated beads predate gate:*)human label → not gatedParked gates — ateam human-list is the canonical needs-human view:
ateam human-list
Issues carrying the human label (set by the gate protocol via ateam gate) appear here. ateam human-list renders enriched output that includes the gate KIND (REVIEW or QUESTION) and the gate note (e.g. the PR URL) per gated initiative — it does not emit raw JSON. Derive the kind from the labels array in ateam list-json (step 1 above) if you need the programmatic form.
Render ONE markdown table, ordered needs-human first. Columns, in order:
| column | content |
|---|---|
| (icon) | status at a glance: ⚠ needs human · ✅ delivered (PR open/awaiting merge) · 🔍 investigating/root-causing · ▶ executing · 📋 planning · ⏸ parked/blocked (non-human). Pick the one that fits the latest phase note; when in doubt use the closest. |
| ID | the initiative id (e.g. at-zot) |
| Title | the initiative title, untruncated unless absurdly long |
| Phase | a SHORT phase token distilled from the latest note (e.g. planning, executing, root-caused, delivered) — NOT a sentence |
| Where | PR #N if a PR exists, else the branch: name |
Keep cells terse — the table is the at-a-glance view, not the full story. Do not add prose commentary per row.
Below the table, for each needs-human initiative ONLY, add one footnote line:
⚠ <id>: PR ready — needs review: <gate note / PR url>⚠ <id>: <parked question(s) from the latest gate note>This is the one place full question/note text belongs — cells stay short.
If nothing is open: say exactly that, one line. (No empty table.)
Workspace health: run ateam audit. The global workspace must contain ONLY initiative-tracking beads + role memories. On a clean result, add a single terse line under the table (e.g. _audit clean · no leaked work beads_). If audit reports leaked work beads (feature/plan/discovery beads that belong in a project repo), surface them under a ⚠ leaked work beads heading so they get cleaned up — this is a recurring drift worth catching every time the dashboard runs.
Read-only: this skill never modifies the registry (the audit is read-only too — it reports, it does not delete).
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub mgt-insurance/agent-teams