v0.13.0

Auto-maintained state.md

A bound task's state.md now materializes and stays fresh on its own — the common bind → /spec (or /slice) → /clear flow produces a complete state.md (docs footer and prose) even when handoff is never run.

Highlights

• New trace state command group owns state freshness portably: trace state check returns a neutral seed/refresh/no-op verdict and idempotently reconciles the docs-manifest footer; trace state reflect stamps the prose-fingerprint marker.
• Footer as a projection of the docs directory — reconciled render-on-read in the CLI and eagerly at the bind seam, so state.md exists with a correct footer the moment a task has a real doc. Idempotent, so board polling causes no mtime churn.
• Claude Stop hook consumes the verdict: when the doc set has drifted from what the prose last reflected, it sends the still-warm agent back once to seed or refresh the prose. No drift → no-op, so ordinary turns never block. All staleness logic stays in the CLI, keeping the brain portable.
• Re-entry as the cross-platform tier — the re-entry manifest now carries a stateFreshness block, so agents on read-only or fire-and-forget platforms still see and repair drift when they bind.

UI

• The task page badges stale state prose with a quiet "docs changed since this was saved" crumb, so unrepaired drift is visible to a human.

Skills / docs

• Renamed the handoff skill to trace-state with dual triggers.
• README documents the state.md freshness tiers.