feat: test-performance board — CI wall-clock, hang events, NO_RUN census

[Jammy2211]

Overview

Development speed tracks test speed, and today knowing what is slow takes a manual archaeology session over CI job logs. This task builds the standing surface: a ⏱ Performance section on the Heart board showing run times of the testing/integration infrastructure (PR smoke gates, unit-test gates, weekly Workspace Smoke, Nightly Release), with history to flag drops, kill-timer/hang events surfaced, and the NO_RUN census with reasons — every actionable row carrying its own one-tap Claude prompt. The Brain board consumes the published block verbatim as a headline row (the existing heart_blockers contract).

Full design rationale with receipts: PyAutoMind/docs/pyautoheart/test_performance_board_assessment.md. Mind prompt: active/test_performance_board.md.

Plan

• Add two cloud-safe Heart checks: ci_timing (Actions-API wall-clock scrape of the tracked gates — durations, queue delay, conclusion mix, cancelled/timed-out disambiguation) and no_run_census (parse each workspace's config/build/no_run.yaml; SLOW/NEEDS_FIX/permanent tiers, marker age, unmeasured-SLOW flagging).
• History via the self-carrying published-artifact roll-forward (previous board.json fetched at render, today appended, capped) — no commit noise.
• Two new dashboard sections + an additive performance block in board.json (schema v2 stays additive); rows carry their own /bug prompts.
• Wire both checks into heart-health.yml's daily cloud run (not the <30s tick).
• Brain board: a "⏱ Test performance" section consuming the published block verbatim; hang events count as attention in the verdict.
• Advisory only: the GREEN/STALE/YELLOW/RED readiness verdict is untouched.

Detailed implementation plan

Affected Repositories

• PyAutoHeart (primary)
• PyAutoBrain (board consumer)
• PyAutoMind (prompt lifecycle + assessment doc, already pushed)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoHeart	main → claude/test-performance-dashboard-y3fdy7	clean
./PyAutoBrain	claude/test-performance-dashboard-y3fdy7 (= main)	clean
./PyAutoMind	claude/test-performance-dashboard-y3fdy7	pushed

Branch: claude/test-performance-dashboard-y3fdy7 (all three repos)

Implementation Steps

1. heart/checks/ci_timing.{sh,py} — per polled repo, gh api repos/<owner>/<name>/actions/runs?per_page=50 (PR runs included); durations from run_started_at→updated_at (never created_at); success-only medians per required workflow (+ performance: extra_workflows in config/repos.yaml: Workspace Smoke, Nightly Release); cancelled runs disambiguated (superseded-by-newer-same-branch = benign; on main or successorless = suspect event; timed_out always an event); per-repo sidecars + an aggregate $HEART_STATE_DIR/ci_timing.json that rolls history forward from the previously published board.json (dedupe by date, capped).
2. heart/checks/no_run_census.{sh,py} — contents-API fetch of config/build/no_run.yaml per workspace-group repo; line-based parse (comments carry the markers; a bare off entry must not crash as a YAML boolean); SLOW rows flagged measured: false when the reason carries no real seconds figure (a bare "1800s cap" mention is not a measurement); 404 recorded honestly as present: false.
3. heart/state.py — aggregate the two new global keys.
4. heart/dashboard.py — "CI wall-clock" and "NO_RUN census" sections (cloud-observed, not in LOCAL_ONLY_FAMILIES; events → FAIL state, drift → WARN, both advisory); Board.performance emitted in to_dict as the performance block; sparkline helper for gate history.
5. config/repos.yaml — performance: block + thresholds: ci_timing: {yellow_factor: 1.5, min_delta_s: 120, history_cap: 30} (drift flags only on ratio AND absolute floor — the profiling-conductor doctrine).
6. .github/workflows/heart-health.yml — run both checks in the cloud-safe step.
7. PyAutoBrain board/_board.py — one Heart board.json read serves blockers + performance; ⏱ Test performance section rendering flagged rows' prompts verbatim; events counted as attention.
8. Tests both sides (stdlib+PyYAML, hermetic fixtures, tenant-firewall-clean).

Key Files

• heart/checks/ci_status.{sh,py} — the API-safe check shape being mirrored
• heart/dashboard.py — the one unified renderer (sections, blockers, html 📋 buttons)
• .github/workflows/heart-health.yml — the daily cloud render + Pages publish
• PyAutoBrain/board/_board.py — fetch_heart_blockers consumer contract + _row chips

Out of scope (phased separately)

• Phase 0 prerequisite (separate PR): draft/bug/pyautoheart/script_timing_baselines_orphaned_and_window_filled.md — orphaned slugs / single-observation windows in script_timing.
• Phase 2 (PyAutoHands): per-script smoke_timings.json from the delegated runner — the standing dataset answering item 4 of draft/research/ci/smoke_timing_and_profiling.md.

Original Prompt

Click to expand starting prompt

Test-performance section on the Heart board — run times, hangs, NO_RUN, with one-tap fix prompts

Type: feature / Target: pyautoheart / Repos: @PyAutoHeart @PyAutoHands @PyAutoBrain
Difficulty: large / Autonomy: supervised / Priority: high / Filed: 2026-08-24

Development speed tracks test speed, and today knowing what is slow takes a manual archaeology session over CI job logs. Build the standing surface: a ⏱ Performance section on the Heart board that always shows the run times of the testing/integration infrastructure — PR smoke gates (*_workspace_test, normal workspaces, HowTo), unit-test gates, weekly workspace-smoke, the nightly release driver, import time — with enough history to flag drops, kill-timer/hang events surfaced, the NO_RUN census with reasons, and a one-tap Claude prompt on every row so "speed this up" is a paste.

Full design rationale, data-plane analysis, chip payloads, and the trap list: docs/pyautoheart/test_performance_board_assessment.md in PyAutoMind.

Acceptance:

• The Heart board page shows, per tracked gate: latest duration, p50/max over the window, trend sparkline, conclusion mix, coverage count — refreshed on the Heart's existing render cadence with no commit noise.
• A hang/kill event (per-script TIMEOUT, unexplained cancelled, aborted summary) appears within one render, red, with a /bug … — <run url> chip.
• Every SLOW/NEEDS_FIX no_run.yaml entry is listed with marker age, reason, and a re-measure-first fix chip; unmeasured SLOW markers are visibly flagged.
• The Brain board carries the headline row consuming the published block verbatim.
• Nothing gates: readiness verdict and badge.json message shape unchanged.

[Jammy2211]

Phase 1 implementation is complete and pushed on claude/test-performance-dashboard-y3fdy7 in all three repos.

PyAutoHeart 9d796c8 (+2380/−4): heart/checks/ci_timing.{sh,py} (Actions wall-clock per tracked gate — success-only medians from run_started_at, PR runs included, cancelled disambiguated superseded-vs-suspect, timed_out always an event; self-carrying history rolled forward from the previously published board.json, capped at 30); heart/checks/no_run_census.{sh,py} (line-based marker parse, SLOW/NEEDS_FIX/permanent tiers, unmeasured-SLOW flagging — a bare "1800s cap" mention does not count as a measurement; 404 recorded honestly as present: false); two new dashboard sections + the additive performance block in board.json with every actionable row carrying its own /bug prompt; performance.extra_workflows (Workspace Smoke, Nightly Release) + thresholds.ci_timing (warn only on ratio ≥1.5 AND ≥120s absolute) in config/repos.yaml; both checks wired into heart-health.yml's daily cloud run — deliberately not the <30s tick. 573 tests pass (+73).

PyAutoBrain 49cd3f0 (+297/−14): the ⏱ Test performance section after Readiness — one board.json read now serves blockers and performance; flagged rows (events → warn gates → unmeasured SLOW) render their prompts verbatim as 📋 chips; hang events count as attention in the verdict, never blocking. 447 tests pass (+5).

PyAutoMind da1e8351/83f5f342: design record docs/pyautoheart/test_performance_board_assessment.md, prompt in active/, task registered.

Remaining before close-out:

• PRs + merge (Heart first so the published performance block exists before the Brain render reads it; the Brain side degrades silently on its absence either way).
• After the Heart merge, the first heart-health.yml run seeds the history (sparklines appear from day 3).
• Follow-ups tracked separately: phase 0 (draft/bug/pyautoheart/script_timing_baselines_orphaned_and_window_filled.md, unissued) and phase 2 (per-script smoke_timings.json from the delegated PyAutoHands runner).

---

Generated by Claude Code

[Jammy2211]

Shipped: PyAutoHeart#164 → d8c125e, PyAutoBrain#261 → 1b79bb3 (both merged 2026-08-24, CI green on every leg — the one red was the tenant firewall on a real repo name in a new test file, fixed in b26d94f). Completion record: PyAutoMind/complete/2026/08/test-performance-board.md (Mind PR PyAutoLabs/PyAutoMind#285). The first heart-health.yml run seeds the performance history; sparklines become meaningful from ~day 3. Follow-ups tracked in the record: phase 0 (script_timing baselines bug prompt, unissued) and phase 2 (per-script smoke_timings.json in the delegated PyAutoHands runner).

---

Generated by Claude Code