report

/audit:report — render the status report

Read ${CLAUDE_PLUGIN_ROOT}/reference/orchestrator.md and ${CLAUDE_PLUGIN_ROOT}/reference/manifest-conventions.md first. Run preflight steps 1–2 only (read-only: no git-root/submodule check, no lock).

Run python3 "${CLAUDE_PLUGIN_ROOT}/scripts/report/render-report.py" <manifestPath> (pass --out-dir <dir> through from $ARGUMENTS when given; artifacts otherwise land next to the manifest) and print the written paths.

Optional AI summary (recommended). Before rendering, compose a 2–4 sentence, plain-language summary of the audit's current state — synthesize the phases' desiredOutcome/summary, notable task outcomes, open bugs, and the rollup (audit-status.py <manifestPath> --json). Write it to a scratch file (e.g. <out-dir or manifest dir>/.audit-summary.txt, via Bash) and add --summary-file <that file> to the render command; the renderer shows it in a Summary box in the HTML, the Markdown, and the printable PDF. It is passed to the renderer only — it does NOT modify the manifest, so /audit:report stays read-only (no lock).

The HTML opens standalone (double-click): in-page text/status filtering, area chips when the plan tags areas (they gate phases, like the status chips, and travel in the shareable #! hash as a=), per-phase collapse, a Save as PDF button (browser print, all phases expanded, either orientation) and a Download .md button — all self-contained (inline CSS + JS, zero network fetches), publishable as a CI artifact. The Usage section adds author chips when the ledger records more than one author — they scope that section's per-author views only, because tasks record no author to filter by, and the page says so — and a Month-by-month table once the ledger spans two calendar months, whose plan columns count the whole project by event month. Never locks, never mutates the manifest.

What the report says about test runs

Earned by a pointer and by nothing else. A subject carries testEvidence once a run has been recorded against it, and a pointer — on a task or on a phase — is what turns this whole surface on: the two filter rows, the phase marks, the drawer's third group and the Markdown twin's column. The tests column itself is earned one step narrower, by a task carrying one. A gate the plan merely declares earns nothing, deliberately: reading declarations instead would put a column of No evidence on every manifest written before the field existed, so a plan pointing at no recorded run renders exactly as it did before any of this shipped, and upgrading the plugin changes not a byte of a report somebody already published.