Report

report — the standardized run manifest

Follow the shared contract in ../README.md. This skill produces the one envelope every QAIA plugin shares: .qaia/reports/<US-ID>/manifest.json, defined by ../../OUTPUT-CONTRACT.md (that path is in the QAIA source repository, not in the installed plugin — the fields it defines are restated wherever this skill needs them). It never invents data — it projects the existing journey artifacts into the common schema so qaia-score, an export, or CI can read any run the same way.

Prerequisite

A generated test book in .qaia/testbooks/<US-ID>/ (with its synthesis.md, coverage-matrix.md, and the 03-design.md / 04-priorities.md checkpoints). If it is absent, say which step is missing and offer testbook-generate — never emit a manifest with guessed counts.

Steps

1. Read the source artifacts, never re-generate them:

   • the .feature files (count scenario blocks, priority/@negative/@smoke/Outline tags, technique tags, @oracle:* provenance);

   total counts executable cases, not blocks. A Scenario Outline with N Examples rows counts N — that is what a runner will execute, and what testbook-export already projects as N rows. outlines carries the number of un-expanded blocks. Counting blocks instead made a book of 10 Outlines × 6 examples read as 10 scenarios in the manifest and 60 rows in the reviewer's spreadsheet, with every ratio — negative, low-confidence, byPriority — computed on the wrong denominator, and aptitude-gate deciding a release on it (2026-08-10). The shipped scorer exposes both under executableCases and scenarios; when it has run, take total from executableCases rather than recounting.

   • coverage-matrix.md (AC covered, condition coverage);
   • 03-design.md ([req-neg] conditions → reqNegTotal);
   • synthesis.md and 02-understanding.md (open questions, assumptions, simulated, low-confidence);
   • 03-design.md and the # rule: BR-KB-nnn scenario comments → design.knowledgeApplied (the knowledge-base rules that shaped the book — the provenance that shows the team's own rules were actually applied, not merely available).

1b. Count the level tags → design.byLevel (contract 1.1, ADR 0008). Closed keys e2e / api, counted on the same denominator as scenarios.total — an Outline with N Examples rows contributes N to its level, exactly as it contributes N to the total. The two must therefore sum to total; the shipped validator rejects a manifest where they do not.

Emit the block only if every scenario carries a level tag. A book generated before 2026-08-11, or hand-edited, may have scenarios without one: in that case omit byLevel entirely and say which scenarios lack the tag. The contract makes it optional precisely so that a partial count is never shipped — a byLevel covering 18 of 22 scenarios would read as a coverage-by-level statement that nobody established, which is worse than its absence.

2. Compute the counts — do not estimate. Every number in the manifest must equal what the artifacts contain: the negative ratio is @negative executable cases / all executable cases — the same denominator as scenarios.total, per the output contract's arbitration of 2026-08-10 and the single definition in testbook-generate/references/negative-ratio.md. (This line said "blocks / all blocks" until 2026-08-11: a third copy of a rule the contract had already settled, found by computing the ratio of a real book rather than re-reading the texts.) reqNegCovered/reqNegTotal is the negative-path coverage gate of ADR 0001, byPriority sums to total minus the excluded @smoke journey per the counting rules of testbook-generate.
3. Merge, don't clobber (contract rule 2). If manifest.json already exists, load it, replace only the design section and openArbitrations, append this skill to producers[], and add any new artifacts[] entries — leaving execution, gate, and a human-set status untouched. On a first-ever write, omit gate entirely (not null) — the schema treats its absence as "not yet scored"; only qaia-score ever writes this key (contract rule: no producer self-scores).
4. Fill openArbitrations from every still-pending ⚠ VALIDATION point: [open] questions, simulated defaults, waivers awaiting confirmation — each with its sourceCheckpoint. A non-interactive run surfaces all its simulated entries here.
5. Write .qaia/reports/<US-ID>/manifest.json (create the directory). On a surface without file tooling, emit the JSON as a fenced block and tell the user where to save it.
6. Report the headline line to the user: US-ID, scenarios by priority, by level, AC coverage, negative-path gate (reqNegCovered/reqNegTotal), open arbitrations count — and remind them the gate verdict is filled by qaia-score, not here.

Guardrails

• Never self-score. This skill writes design, artifacts, openArbitrations, provenance — never the gate block (shared contract rule 3: the skill never self-validates; scoring is a separate plugin).
• No secrets, no PII, no raw source in the manifest — counts, IDs, paths, verdicts only (contract principle 5). If a title or path would leak sensitive data, redact it.
• Counts must match the book. If the manifest and the artifacts disagree, the artifacts win: stop and surface the discrepancy rather than writing numbers that lie (same rule as testbook-export).
• Contract version. Always stamp "contract": "1.0"; if ../../OUTPUT-CONTRACT.md has advanced, follow its current version and note the bump.