| name | browser-qa-fix |
| description | [Tier 2 · moderate autonomy · full review gate · browser-grounded] Systematically QA a web application via real browser sessions, freeze findings with screenshot EVID, then fix and re-verify in a loop until health score meets threshold. Maps gstack /qa + /browse + /qa-only into fleet worktrees with build-blind review. Use when a staging or local URL exists and the user wants "test the site and fix bugs", "browser QA", or "dogfood this flow". Trigger on: "browser qa fix", "test the site and fix", "QA this app", "find UI bugs and fix them", "run QA on staging". |
| license | MIT |
| compatibility | Requires git and gh CLI; target app reachable at a URL (local or staging) |
| metadata | {"author":"ravidsrk","version":"1.0.0","tier":"2","fleet-component":"mission","recommended-bundle":"gstack-browser","community-recommends":{"bundle":"gstack-browser","skills":["browse","qa","qa-only"],"mode":"warn"}} |
| status | exploratory |
Status: exploratory. Derived from gstack qa, qa-only, and browse; not yet proven on
an external repo archive. See docs/exploratory/missions/README.md.
Mission: browser-qa-fix
Required skills
Before executing, activate these skills and read their full instructions:
autonomous-fleet-core — read references/engine.md, references/composition.md, EVID rules- One runtime adapter:
autonomous-fleet-adapter-orca, autonomous-fleet-adapter-claude-code,
autonomous-fleet-adapter-grok, or autonomous-fleet-adapter-codex
bundle: gstack-browser
skills:
- browse
- qa
- qa-only
mode: warn
Optional skills
| Skill | Activate when | If unavailable |
|---|
browse / gstack-browse (gstack) | Headless $B or gstack browse daemon available | Playwright CLI / browser skill per adapter |
qa-only (gstack) | User wants report-only pass first | T-QA-SCAN still runs; fix loop skipped until user confirms |
design-review (gstack) | Visual/regression sampling after fixes | Screenshot diff in EVID only |
health (gstack) | User wants composite health scorecard | Mission before/after health fields in readiness doc |
Community catalog: autonomous-fleet-core → references/community-skills.md. At most 2 optional
skills active.
Worker skills
| Role | Skills | If unavailable |
|---|
| @grok (browser repro + fix) | browser, qa (gstack) when browse stack installed | Adapter browser tools + manual repro steps |
| @claude (QA plan, blind review) | — | Mission review gate |
| @claude (integrator) | — | Mission ship gate |
Builder choice: @grok retained for browser/visual fix loops (gstack /qa precedent). Fresh
build-blind @claude reviews each fix PR.
Deferred missions
Record in <LEDGER_DIR>/qa-readiness.md under Recommended next missions.
| Finding type | Route to |
|---|
| Backend-only bug (no browser repro) | bug-batch (exploratory) |
| Systemic architecture issue | adversarial-review-and-fix (shipped) |
| Missing test coverage for fixed area | test-coverage (shipped) |
| Full design parity drift | design-integration (exploratory) |
TARGET URL (user supplies)
Staging URL, preview deploy URL, or http://localhost:<port> with explicit start command recorded
in <LEDGER_DIR>/qa-progress.md. HARD EXTERNAL DEPENDENCY if app cannot be reached — pause and document.
GOAL
Drive a web app from known-broken or unknown to QA health threshold met with every
confirmed defect closed via a PR, each with browser EVID (screenshot + steps), and a before/after
health summary.
ROLE PIPELINE
- @claude compiles user-story checklist and QA tier (Quick / Standard / Exhaustive per gstack
/qa semantics).
- @grok (or coordinator with browse tools) executes browser repro for each story; failures →
frozen
docs/qa-index.md rows with EVID paths under .fleet/evidence/ or docs/qa-evidence/.
- @grok fixes one defect per PR; attaches EVID after-state.
- Fresh build-blind @claude reviews each fix (repro replay + regression risk).
- @claude integrates and updates health scores in ledger.
LEDGER
<LEDGER_DIR>/qa-progress.md — per-story flags + rolling health score (before/after).
Frozen index: docs/qa-index.md — columns: id, severity, story, repro_steps, evid_before,
fix_pr, evid_after, status: OPEN|FIXED|WONTFIX.
EVID GATE
Every OPEN row in qa-index must have evid_before with: URL, viewport, steps, screenshot path,
timestamp. FIXED requires evid_after replaying same steps. Screenshot existence alone is not
sufficient — steps must be copy-pasteable.
TASK STRUCTURE
- T-QA-PLAN [@claude] — ingest URL + tier; write story checklist to
<LEDGER_DIR>/qa-progress.md;
seed docs/qa-index.md.
- T-QA-SCAN [@grok + browse] — run stories; record failures with EVID; compute health_before.
- T-FIX-LOOP [per qa-index OPEN row] — @grok fix → fresh @claude review → @claude ship →
update row FIXED + evid_after.
- T-FINAL [@claude] — re-run full scan at chosen tier; health_after must meet threshold
(Quick: no critical/high open; Standard: no medium+; Exhaustive: cosmetic catalogued only).
Output
<LEDGER_DIR>/qa-readiness.md with fleet-outcome YAML (health_before, health_after,
defects_open, defects_fixed).
Runtime goal
Mission browser-qa-fix DONE: <LEDGER_DIR>/qa-progress.md complete, <LEDGER_DIR>/qa-readiness.md with
fleet-outcome.status done, health_after meets tier threshold,
./scripts/validate-fleet-outcome.sh passes.
DONE
All in-scope qa-index rows FIXED or WONTFIX with documented rationale, health threshold met,
all fix PRs merged, <LEDGER_DIR>/qa-readiness.md with fleet-outcome present.
DECISION DEFAULTS (mission-specific)
- Browser repro beats log-only claims; no fix without EVID before-state.
- One defect per PR; atomic commits per gstack
/qa convention.
- Flaky repro → mark
FLAKY in index, do not fix until repro stable.
- Cosmetic defects deferred in Quick tier; included in Exhaustive.
- If browse daemon unavailable, fall back to Playwright with same EVID schema.
