swain-status（swain 状态）

Status

Cross-cutting project status dashboard. Aggregates data from artifact lifecycle (swain chart), task tracking (tk), git, GitHub issues, and session state into an activity-oriented view.

Roadmap freshness

The status script includes a staleness check that regenerates ROADMAP.md if it is missing or older than any doc artifact. This runs automatically — no separate invocation needed.

For a full roadmap refresh (unconditional regeneration), use swain-roadmap instead.

When invoked

Locate and run the status script from skills/swain-status/scripts/swain-status.sh:

# Find the script from the project root or installed skills directories
SKILL_DIR="$(find . .claude .agents -path '*/swain-status/scripts/swain-status.sh' -print -quit 2>/dev/null)"
bash "$SKILL_DIR" --refresh

If the path search fails, glob for **/swain-status/scripts/swain-status.sh.

The script's terminal output uses OSC 8 hyperlinks for clickable artifact links. Let the terminal output scroll by — it is reference data, not the primary output.

After the script runs, present a structured agent summary following the template in references/agent-summary-template.md. The agent summary is what the user reads for decision-making. It must lead with a Recommendation section (see below), then Decisions Needed, then Work Ready to Start, then reference data — following the template in references/agent-summary-template.md.

The script collects from five data sources:

1. Artifacts — swain chart vision-rooted hierarchy (epic progress, ready/blocked items, dependency info). Use bash "$(find "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" -path '*/swain-design/scripts/chart.sh' -print -quit 2>/dev/null)" recommend for ranked artifact view; respects focus lane automatically.
2. Tasks — tk (in-progress, recently completed)
3. Git — branch, working tree state, recent commits
4. GitHub — open issues, issues assigned to the user
5. Session — bookmarks and context from swain-session

Compact mode (MOTD integration)

The script supports --compact for consumption by swain-stage's MOTD panel:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
STATUS_SCRIPT="$(find "$REPO_ROOT" -path '*/swain-status/scripts/swain-status.sh' -print -quit 2>/dev/null)"
[ -n "$STATUS_SCRIPT" ] && bash "$STATUS_SCRIPT" --compact || echo "swain-status.sh not found"

This outputs 4-5 lines suitable for the MOTD box: branch, active epic progress, current task, ready count, assigned issue count.

Cache

The script writes a JSON cache to the project-local agents directory:

.agents/status-cache.json

• TTL: 120 seconds (configurable via status.cacheTTL in settings)
• Force refresh: --refresh flag bypasses TTL
• JSON access: --json flag outputs raw cache for programmatic use

The MOTD can read this cache cheaply between full refreshes.

Migration: If .agents/status-cache.json does not exist but ~/.claude/projects/<project-path-slug>/memory/status-cache.json does, read the old location once and write to the new location going forward.

Recommendation

The recommendation uses a scoring formula:

score = unblock_count × vision_weight

Where vision_weight is inherited from the artifact's Vision ancestor (high=3, medium=2, low=1, default=medium). Read .priority.recommendations[0] from the JSON cache for the top-ranked item.

Tiebreakers (applied in order):

1. Higher decision debt in the artifact's vision
2. Decision-type artifacts over implementation-type
3. Artifact ID (deterministic fallback)

When a focus lane is set, recommendations are scoped to that vision/initiative. Peripheral visions are summarized separately (see Peripheral Awareness).

If attention drift is detected for the recommended item's vision, include it as context in the recommendation.

Mode Inference

swain-status infers the operating mode from context (first match wins):

1. Both specs in review AND strategic decisions pending → ask: "Steering or reviewing?"
2. Specs awaiting operator review (agent finished, needs sign-off) → detail mode
3. Focus lane set with pending strategic decisions → vision mode within that lane
4. No specs in review, decisions piling up → vision mode
5. Nothing actionable in either mode → vision mode (show the master plan mirror)

Once the operator answers, swain remembers for the session via swain-session bookmark.

Focus Lane

The operator can set a focus lane via swain-session to scope recommendations to a single vision or initiative:

bash "$(find . .claude .agents -path '*/swain-session/scripts/swain-focus.sh' -print -quit 2>/dev/null)" set VISION-001

When set, .priority.recommendations only includes items under that vision. Non-focus visions appear in the Peripheral Awareness section.

Peripheral Awareness

When a focus lane is set, non-focus visions with pending decisions are summarized: "Meanwhile: [Vision] has N pending decisions (weight: W)"

This is a mirror, not a recommendation — the operator decides when accumulation warrants redirection.

Active epics with all specs resolved

When an Active epic has progress.done == progress.total:

• Show "→ ready to close" in the Readiness column of the Epic Progress table
• Do NOT show it in the Work Ready to Start bucket (it's not implementation work)
• Do NOT show it as "work on child specs"

Decisions Needed (roadmap integration)

After the existing status output sections, surface top decision items from the roadmap. This section uses the Eisenhower classification from chart.sh roadmap --json.

Data collection

bash "$(find . -path '*/swain-design/scripts/chart.sh' -print -quit)" roadmap --json

Filtering

1. Focus lane scoping: If a focus lane is set, filter items to that Vision only.
2. Eisenhower quadrant filter: Show items from "Do First" and "Schedule" quadrants. Classification criteria:

- Important: weight >= 3 - Urgent: active status or score > 0

1. Operator decision filter: Show items that need operator decisions:

- Status "Proposed" — needs activate or drop decision - children_total == 0 — needs decomposition into child specs - children_complete == children_total and children_total > 0 — ready to complete/close

Display

• Limit to top 5 items, ordered by weight descending then score descending.
• Format each as an actionable prompt, not a passive list entry. Examples:

- EPIC-038 PR-Only Agent Guardrails — **needs decomposition** (0 specs, high priority) - EPIC-024 GitHub Issue Polling — **activate or drop?** (Proposed, high priority) - EPIC-012 Session Atomization — **ready to complete** (4/4 specs done)

• If no items match, omit the section entirely — do not show an empty heading.

Degradation

If chart.sh roadmap --json fails or returns empty data, skip the Decisions Needed section silently and continue with the rest of the status output.

Settings

Read from swain.settings.json (project) and ~/.config/swain/settings.json (user override).

Session bookmark

After presenting status, update the bookmark with the most actionable highlight: bash "$(find..claude.agents -path '*/swain-session/scripts/swain-bookmark.sh' -print -quit 2>/dev/null)" "Checked status — {key highlight}"

Error handling

• If chart.sh / specgraph is unavailable: skip artifact section, show other data
• If tk is unavailable: skip task section
• If gh CLI is unavailable or no GitHub remote: skip issues section
• If .agents/session.json doesn't exist: skip bookmark
• Never fail hard — show whatever data is available