status skill

setup-status

Render a one-glance picture of how apache-magpie is adopted in this repo, then let the user reconfigure it without leaving the view. The dashboard answers the questions an operator actually asks: which install method and version are pinned, has the snapshot drifted, which agent targets are wired, which skill families are installed, and are the symlinks healthy.

This skill is the configuration view of adoption. It is read- only on its own; every change it offers is carried out by delegating to setup — the one skill that owns adoption mutation. For a deep integrity / health check (lock parsing, per-check ✓/✗ matrix, permission-hygiene audit, ASF comdev MCP prerequisites, stale-worktree sweep), use setup verify; this skill links to it rather than duplicating its checks.

---

Adopter overrides

Before running its default behaviour, this skill consults setup-status.md in the personal layer (.apache-magpie-local/ when the project adopted Magpie, falling back to the main checkout's in a linked worktree, or <git-common-dir>/apache-magpie/ when Magpie is only installed; applied first, wins on conflict) and .apache-magpie-overrides/setup-status.md (committed, project-wide) in the adopter repo, if present, and applies any agent-readable overrides it finds. See docs/setup/agentic-overrides.md for the contract.

Hard rule: agents NEVER modify the snapshot under <adopter-repo>/.apache-magpie/. Local modifications go in the override file; framework changes go via PR to apache/magpie.

---

Snapshot drift

Also at the top of every run, this skill compares the gitignored .apache-magpie.local.lock (per-machine fetch) against the committed .apache-magpie.lock (the project pin). On mismatch the skill surfaces the gap and proposes setup update. The proposal is non-blocking; surfacing the drift is itself part of this skill's dashboard, so the comparison feeds Step 1 rather than gating the run.

---

Inputs

Skill directives (how the user invokes the skill):

Input	Effect
(none)	Render the dashboard, then offer adjustments interactively.
--no-adjust	Render only; skip the reconfigure offer.
adjust	Skip straight to the reconfigure flow after a brief state recap.

Collector flags (passed through to scripts/collect_status.py):

Flag	Effect
--repo <path>	Inspect a repo other than the current git top-level.
--format md	The Markdown dashboard (default).
--format json	The raw machine-readable fields (collect.md); add --pretty to indent.

The default output is the Markdown dashboard, rendered deterministically by the collector.

---

Prerequisites

• A git checkout (git rev-parse --show-toplevel succeeds), or an explicit --repo <path>.
• python3 on PATH for the deterministic collector (scripts/collect_status.py). No third-party packages, no network access.

This skill reads only framework-internal, on-disk state (lock files, symlinks, .gitignore, the post-checkout hook). It reads no external or private content, so the prompt-injection and Privacy-LLM gate-checks do not apply.

---

Step 0 — Pre-flight check

1. Resolve the repo root (--repo if given, else git rev-parse --show-toplevel).
2. Apply adopter overrides and the drift preamble above.
3. Adopted? If <committed-lock> (.apache-magpie.lock) is absent, the repo is not adopted. Say so, point at setup to adopt, and stop — there is no state to render.

---

Step 1 — Render the dashboard

The collector renders the dashboard itself. Run it (it never writes and never fetches):

python3 <framework>/skills/setup-status/scripts/collect_status.py

(From a normal adopter the script lives under the snapshot at .apache-magpie/skills/setup-status/scripts/; invoke it via the magpie-setup-status symlink's resolved path. The default output is the Markdown dashboard; pass --format json only when tooling needs the raw fields — see collect.md.)

OUTPUT CONTRACT — non-negotiable. Present the script's Markdown output verbatim (it is already GitHub-flavoured Markdown — let the harness render the pipe table). Do not: re-draw it as a box-drawing/ASCII table; drop the serves bullet legend (it carries the agents each directory serves, including the whole universal cluster); add a Reads column back into the table (that is what made it wrap and break); recompute the verdict; or "prettify" the layout. The script, not the agent, owns the rendering, precisely because an LLM formatting pass reliably mangles it (drops the agents-served legend, renames columns, re-introduces the wide column). If you find yourself rebuilding the table, stop and paste the script output instead.

The renderer owns the headline, the agent-target table plus its serves legend (so the universal cluster and every registry vendor always appear), the family roster, and the drift / integrity summary.

Full layout reference, the health-verdict rules, and mode-aware interpretation: render.md.

---

Step 2 — Interpret (lightly)

After printing the verbatim dashboard, optionally add a one-line mode-aware note where it helps — without re-tabulating or contradicting the script output. Example: a method:local framework checkout commits its symlinks and has no snapshot or local lock, so the absent snapshot is healthy there but a fault for a normal adopter (render.md).

---

Step 3 — Offer adjustments

Unless --no-adjust was passed, end the dashboard with the reconfigure offer. Detect the obvious deltas (a registry target present on disk but not wired; an opt-in family not installed; dangling symlinks; drift) and present each as a concrete, confirmable change. On confirmation, delegate to setup with the right flags (agents:<list>, skill-families:<list>, or update) — this skill never edits symlinks, locks, or .gitignore itself.

Full gap-detection and delegation rules: adjust.md.

---

Hard rules

• Read-only on its own; mutation only via setup. This skill never writes a symlink, a lock file, or .gitignore. Every change is delegated to setup, preserving the framework's single source of truth for adoption mutation.
• Propose before applying. Each adjustment is a proposal the user explicitly confirms before the delegated setup run starts. No silent reconfiguration.
• Do not duplicate verify. For deep integrity, permission hygiene, comdev-MCP prerequisites, and the stale-worktree sweep, point the user at setup verify rather than re-implementing those checks here.
• Never invent state. Report only what the collector observed. If a field is unknown (e.g. upstream-tip drift needs network), say it was not checked and name the skill that does check it.

---

References

• collect.md — the collector's JSON field reference.
• render.md — dashboard layout, health-verdict rules, mode-aware interpretation.
• adjust.md — gap detection and the delegation contract to setup.
• scripts/collect_status.py — the deterministic, read-only state collector.
• setup — adoption mutation: install.md, update.md, the agents.md target registry, and the Golden rules.
• setup verify — the deep integrity / health check this dashboard complements.
• AGENTS.md — framework conventions and the placeholder convention.
• docs/setup/agentic-overrides.md — the override contract every skill consults.