qa-feature-catalog

name	qa-feature-catalog
description	Maintain a semantic QA feature catalog for Harnessy-compatible repositories by consolidating raw spec suites into stable feature slugs, prefixes, result files, and optional result-sink tabs.
disable-model-invocation	false
allowed-tools	Read, Write, Grep, Glob, Bash, ApplyPatch
argument-hint	[--dry-run] [--write-overrides] [--features <slug,...>] [--apps <app,...>]

QA Feature Catalog

Purpose

Keep a repo's QA feature catalog stable and human-meaningful while preserving deterministic runtime behavior. The catalog maps scenario ID prefixes to feature slugs, display names, apps, regression suites, result snapshots, and optional external result sinks.

Template paths are resolved from ${AGENTS_SKILLS_ROOT}/qa-feature-catalog/.

Inputs

• Canonical QA profile and parsed qa ids output.
• Optional catalog artifacts:
  • qa/features.generated.yaml
  • qa/features.overrides.yaml
  • qa/features.changelog.md
• Optional result snapshots under qa/run-results/.
• Optional result-sink metadata from the QA profile or delivery profile.

Steps

1. Follow ${AGENTS_SKILLS_ROOT}/qa-feature-catalog/commands/qa-feature-catalog.md.
2. Parse all specs with qa ids --json.
3. Group scenarios by ID prefix, app, layer, suite annotation, and spec source.
4. Compare the deterministic grouping to existing generated catalog and overrides.
5. Propose semantic decisions: accept, rename, merge, split, archive, or add.
6. When --write-overrides is requested, update only the hand-maintained override/changelog artifacts. Do not hand-edit generated catalog files owned by repo-local generators.
7. Run the repo-local catalog generator/validator when available, then run qa drift.

Output

• Decision table with each proposed feature catalog change.
• Optional updates to qa/features.overrides.yaml and qa/features.changelog.md.
• Instructions for any repo-local generated files or external result sinks that need migration.

Guardrails

• Do not rename an ID prefix that already has committed results or external rows without an explicit migration plan.
• Do not invent features with no spec backing unless they are explicit manual or cross-feature entries.
• Keep generated files generated. Use overrides for semantic tie-breaking.
• Preserve existing result snapshots and note any orphaned rows.