feat(status): area-coverage summary — every reconciled area, not "mostly skills"

[alex-mextner]

What

rig status rendered only the items that were drifting, as one flat dump dominated by skill rows (one per skill + a harness-link row each). Every other area — agent-hooks, the git-hook dispatcher, CI gates, the ship gate, MCP, AGENTS/CLAUDE symlinks, repo settings, harness, tmux, the model-freshness cron, the tg-ctl daemon — was buried under the skill lines or, when in sync, invisible. A user could not see the full picture of what rig manages and where it's out of sync.

ROADMAP: "rig status must cover ALL reconciled areas, not mostly skills" — group by area + the GLOBAL/REPO layer split; each area shows in-sync vs drift.

Note: the sibling ROADMAP items in the same heading cluster — the non-git handling ("non-git dir must NOT claim 'no rig.yaml, should be committed'") and the loud "extras are left for you to decide" reassurance in status + apply --help — already landed in #27 (error-system v2). This PR closes the remaining, biggest gap: the area-coverage summary. I left the already-done work untouched.

How

Add an AREA SUMMARY headline at the top of rig status: every reconciled area, grouped by the GLOBAL/REPO layer, each with its state — not configured / in sync / drift (N declared-but-missing/modified[, M on-disk-not-declared]). The detailed per-item drift dump still follows as the detail.

• riglib/areas.py — the single source of truth for the area list (an Area registry + layer split), mirroring layers.py. The ci category splits into two user-facing areas — CI gates and ship / gh ship — by action slot / drift item, direction-aware so a disk→config extra workflow (even a rogue ship.yml) routes to CI gates, never the ~/bin ship merge gate.
• Counts are the unit-stable DRIFT counts; no fragile declared − missing arithmetic (a plan action and a drift item are not 1:1 across areas — gitignore is one action but two checks, a skill is a copy + a harness-link).
• tg_ctl off Darwin renders unsupported (not a false in sync) — only when configured and genuinely no drift, so a real off-Darwin drift can never be masked.
• layers.py — add gitignore + tg_ctl to _CATEGORY_LAYER (the registry was meant to be exhaustive; both previously fell through to the GLOBAL default).
• _LAYER_HEADERS lifted to a module constant, shared with the per-item drift dump (one source of wording).

Acceptance evidence

rig status against a real agent-tools checkout — all 13 areas visible at a glance, grouped by layer, each with its honest state:

rig status
  repo: /private/var/folders/1c/d4v7_mrs4b9085f6w7w9p35c0000gn/T/tmp4y_pjvqk/repo
  stack: unknown  type: unknown
  config layers: repo:/private/var/folders/1c/d4v7_mrs4b9085f6w7w9p35c0000gn/T/tmp4y_pjvqk/repo/rig.yaml

  areas rig manages:
    GLOBAL — machine-wide (from ~/.config/rig/config.yaml)
      skills: drift (76 declared-but-missing/modified)
      agent-hooks (v1 descriptors): drift (9 declared-but-missing/modified)
      git-hooks dispatcher: not configured
      global gitignore excludes: drift (2 declared-but-missing/modified)
      MCP servers: not configured
      harness auto-mode + hook bridge: not configured
      tmux config: not configured
      model-freshness cron: not configured
      tg-ctl inbound daemon: drift (1 declared-but-missing/modified)
    REPO — this repository (from ./rig.yaml)
      CI gates: drift (1 declared-but-missing/modified)
      ship / `gh ship` merge gate: drift (1 declared-but-missing/modified)
      AGENTS.md / CLAUDE.md symlinks: drift (1 declared-but-missing/modified)
      repo settings (branch protection / GHAS / merge): in sync

  [GLOBAL] tg-ctl inbound daemon: drifted (tg-ctl boot LaunchAgent not installed)  (launchd boot agent, 'ai.hyperide.tg-ctl')

The old output would dump 88 drift lines with skills: dominating; the new summary surfaces every area (incl. repo settings: in sync, the once-buried ship, agent-hooks: drift (9)) in 16 lines.

Tests

riglib/areas.py registry: exhaustive vs the ROADMAP areas, layer partition, ci/ship split on action + drift + direction, frozen dataclass, every known category maps to an area / every area category is layer-classified (guards against a future category silently going uncounted — the headline/detail divergence this feature exists to prevent).

End-to-end status (drives main(["status", …])): all areas listed, drift counts, in-sync marking, partial drift, modified-direction counting, ci-extra + rogue-ship.yml routing, harness action↔drift coherence (no false in-sync for the merged auto-mode + hook-bridge area), gitignore one-action-two-checks in-sync, tg_ctl unsupported/disabled off Darwin, non-git GLOBAL-only.

...............................sssss                                     [100%]
679 passed, 5 skipped in 5.18s

Full suite + tests/smoke.sh (the CI gate: --help/doctor/init/apply-idempotency/status/non-git/tg-ctl + pytest) green. Reviewed across multiple review --staged rounds (Fable/Opus/Codex/GLM), all [ok]; every actionable correctness finding addressed.

Deferred (follow-ups, not regressions)

These are pre-existing detection-contract semantics the summary faithfully reflects (the old status surfaced nothing for them either); fixing them couples the summary to area-specific detect state or surfacing plan.notes, out of scope here:

• A git repo with no GitHub remote shows repo settings: in sync even though detect() intentionally checks nothing (no remote → no drift). Should read not applicable / no GitHub remote.
• An enabled-but-unprovisionable area that the plan builder skips with a plan.notes entry (e.g. models with a missing checker_path, or hook-bridge on an old agent-tools checkout) shows not configured / in sync, hiding the skip. Status should surface skipped/blocked plan notes.
• tg_ctl.boot: false shows in sync in the summary while the dedicated detail line says disabled.

🤖 Generated with Claude Code