board: every evidence gap carries the check that closes it

[Jammy2211]

A STALE board offered nothing copyable that closed a gap: every stale reason got the same /health re-run the stale evidence: <text> string, and a reason row carried no command on any surface (commands existed only on section rows). STALE is the board's steady state — STALE · score 65 on every heart-health.yml run since 2026-08-19 — so that is most of what the board is asked to do.

What changed

• heart/readiness.py — compute emits stale_details: each stale reason with the gate key that produced it (install_unknown, validation_absent, …), index for index with stale_reasons. Additive; the flat reason lists, the verdict, the score, the profiles and the release gate are untouched (tests/test_readiness.py passes unmodified). Internally the stale.append(...) / hit(...) pairs became one add_stale(msg, key) helper, so a gap cannot be filed without its key or its penalty.
• heart/dashboard.py — STALE_REMEDIES maps that key to the check that closes the gap. A stale row gains command (the shell remedy, None where the remedy needs a conversation) and a prompt that names both the check and the gap. The tier gains stale_plan: one prompt walking every current gap, plus the shell chain that does the same — offered only when every gap has a command, since a chain that silently skips one reads as if it cleared it.
• heart/fix.py + bin/pyauto-heart — pyauto-heart fix stale: the same payloads in a terminal, from the same persisted verdict, so the two surfaces cannot prescribe different things. readiness now ends a stale block with → clear them: pyauto-heart fix stale.

Per surface: html renders a ⌨ command chip beside the 📋 prompt chip and a "clear them all" line above the gaps; md leads its prompts block with the plan; json carries blockers[].command and stale_plan (schema v3, additive).

Invariants held

• Every remedy re-runs a check; none touches code. A stale row can never emit a /bug prompt — asserted in the tests.
• Remedies are looked up by gate key, never sniffed from the reason text. A reason with no key (a verdict from an older Heart) falls back to the generic nudge, and no plan is offered, rather than a guessed remedy.
• Observer-only: the Heart still never dispatches a rehearsal, so the release-validation gaps carry a prompt (/release rehearse) and no command.

Deliberately not done

• The README strip stays one line for STALE — test_md_brief_is_one_line_unless_something_is_wrong pins that on purpose, and a glance surface is the wrong place for a multi-step plan.
• skew_pypi_unknown carries a prompt but no command: the deep PyPI leg has no pyauto-heart verb, and inventing one would have been a fake remedy.

Testing

pytest tests/ -q → 615 passed (8 new dashboard tests, 3 new readiness tests). Rendered a synthetic cloud snapshot end to end: it reproduces the live STALE · score 65 with the three current gaps, and pyauto-heart fix stale prints their commands and the one plan.

The Brain-side consumer is PyAutoLabs/PyAutoBrain#(companion branch claude/pyautoheart-stale-r4swxg); a Brain board reading an older Heart surface renders exactly as before.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WY6Q2fZquoDQijLD6PXtAT

---

Generated by Claude Code