When this skill is invoked, orchestrate the QA team through a structured testing cycle.
Decision Points: At each phase transition, use AskUserQuestion to present
the user with the subagent's proposals as selectable options. Write the agent's
full analysis in conversation, then capture the decision with concise labels.
In collaborative mode, the user must approve before moving to the next phase.
In guided mode the pipeline advances automatically unless a phase is BLOCKED;
in autonomous mode it runs end to end, recording each phase outcome via
log_decision. Decisions in automation_always_ask categories
(is_always_ask_category helper) always prompt regardless of mode. See
.claude/docs/automation-modes.md.
Phase 0: Resolve Config
!bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys review_mode,automation,team.size
Resolved above — use as-is; --review overrides review_mode. No block →
defaults in .claude/docs/config-resolution.md.
review_mode sets gate depth:
full— spawn all director and lead gates as describedlean— skip director gates unless they are PHASE-GATE type (CD-PHASE-GATE, TD-PHASE-GATE, PR-PHASE-GATE, AD-PHASE-GATE)solo— skip all director gate spawning entirely; run the skill without any agent gates
automation drives the Decision Points note above. See the Decision Points note above and
.claude/docs/automation-modes.md for how each mode changes pipeline behavior.
team.size: which agents are active (orthogonal to review_mode gate-depth and workflow docs).
individual(default):qa-testeronly;qa-leadinvoked at phase gates only.small:qa-lead+qa-testerpipeline (as documented).studio:qa-lead+ per-storyqa-testerspawn + sign-off. Directors (CD/TD/PR) still spawn at phase gates regardless of size; a non-core agent needed atindividualroutes through the nearest active core agent with an informational note. "Phase gate" means any phase that ends in anAskUserQuestiondecision point before the pipeline advances — not every phase. Apply the test literally: if the phase below has no decision point, it is not a gate, and an agent restricted to "phase gates only" is not spawned for it. This active-set scoping applies throughout the pipeline below: any phase that names an agent outside the active set routes through the nearest core agent rather than spawning it.
Announce the active set before Phase 1 — never let the collapse be silent.
Before spawning anything, state in one line which agents this run will actually
spawn, and which the pipeline below names but will not spawn at the resolved
team.size. For example:
Active set (team.size: <resolved>): <the agents listed for that size above>.Not spawned this run: <every other agent this pipeline names> — consultedthrough <nearest active core agent>. Raise team.size (or modes.rigor) to widen.
Fill it from the team.size list directly above and the agents this file's own
pipeline names — not from an example. Both sets differ per orchestrator.
The pipeline below reads as a multi-agent fan-out and at the shipped default it
is one or two agents — team-release names eight and runs one, team-narrative
names six across five phases and runs writer alone. The collapse is correct:
team.size is rigor-fronted and the narrow default is the token lever, measured
at roughly 10x. What was wrong is that nothing said so, so a reader could not
distinguish a correctly-collapsed run from a broken pipeline, and the per-agent
"routes through the nearest core agent with an informational note" rule above
fires at routing time and never states the shape of the run as a whole.
This is the same rule as the skipped-check reporting elsewhere in this file: a constraint that is enforced but never surfaced is indistinguishable, to the person reading the output, from one that was never enforced.
Team Composition
- qa-lead — QA strategy, test plan generation, story classification, sign-off report
- qa-tester — Test case writing, bug report writing, manual QA documentation
How to Delegate
Use the Agent tool to spawn each team member as a subagent:
subagent_type: qa-lead— Strategy, planning, classification, sign-offsubagent_type: qa-tester— Test case writing and bug report writing
Brief each agent — do not dump context. Read the shared inputs once and pass a distilled brief inline: the lines each agent actually needs, never a file path for a document you have already read (an agent handed a path re-reads the whole file). Pass a path only for a document you have not read and only that agent needs.
End every agent prompt with a return contract: "Write your full output to [path] — that named path is your write authorisation under the bounded exception below, so write it without a separate approval prompt. Return only (1) the path written, (2) a ≤5-bullet summary of decisions, (3) any BLOCKED/CONCERNS items, one line each. Do not restate the documents you read." Without it, an agent returns everything it read back into this session.
Why this does not violate the Collaboration Protocol.
CLAUDE.mdrequires an agent to ask "May I write this to [filepath]?" before Write/Edit. A subagent spawned here writes without asking, and that is a deliberate, bounded exception rather than an oversight — the same call already made forconsistency-checkappending toactive.md. The exception holds only when all three are true: (1) the path is one you named in the prompt, so the user approved the destination when they approved the phase; (2) it is a new artifact underproduction/,docs/ortests/, never an edit to existing source or config; (3) the phase that produced it is itself gated by anAskUserQuestionbefore the pipeline advances. Outside those three, the agent must ask. Do not "fix" this by asking per subagent — a prompt per agent per phase makes an orchestrator unusable, which is why the exception exists.
Launch independent qa-tester tasks in parallel where possible (e.g., multiple stories in Phase 5 can be scaffolded simultaneously).
Pipeline
Phase 1: Load Context
Before doing anything else, gather the full scope:
-
Detect the current sprint or feature scope from the argument:
- If argument is a sprint identifier (e.g.,
sprint-03): Globproduction/sprints/for files matching*[sprint-identifier]*.md. Read the matched file. If multiple match, use the most recently modified. - If argument is
feature: [system-name]: glob story files tagged for that system - If no argument: read
production/session-state/active.mdandproduction/sprint-status.yaml(if present) to infer the active sprint
- If argument is a sprint identifier (e.g.,
-
Read
project.stagefromproject.yaml(fallbackproduction/stage.txt) to confirm the current project phase. -
Count stories found and report to the user:
"QA cycle starting for [sprint/feature]. Found [N] stories. Current stage: [stage]. Ready to begin QA strategy?"
Phase 2: QA Strategy (qa-lead)
Spawn qa-lead via Agent to review all in-scope stories and produce a QA strategy.
Prompt the qa-lead to:
-
Read each story file
-
Classify each story by type: Logic / Integration / Visual/Feel / UI / Config/Data
-
Identify which stories require automated test evidence vs. manual QA
-
Flag any stories with missing acceptance criteria or missing test evidence that would block QA
-
Estimate manual QA effort (number of test sessions needed)
-
Before assessing smoke status, check for an existing smoke check report: Glob
production/qa/smoke-*.mdand read the most recently modified file (if found). If a report exists, use its verdict and findings directly — do not re-interview the user. If no report exists, note: "No prior smoke check report found — run/smoke-check sprintbefore proceeding." and set smoke check status to UNKNOWN (treat as PASS WITH WARNINGS for the purpose of continuing). Produce a smoke check verdict: PASS / PASS WITH WARNINGS [list] / FAIL [list of failures] / UNKNOWN (no report found) -
Produce a strategy summary table and smoke check result:
Story Type Automated Required Manual Required Blocker? Smoke Check: [PASS / PASS WITH WARNINGS / FAIL / UNKNOWN] — [source:
production/qa/smoke-[date].mdor "no report found"] — [details if not PASS]
If the smoke check result is FAIL, the qa-lead must list the failures prominently. QA cannot proceed past the strategy phase with a failed smoke check.
Present the qa-lead's full strategy to the user, then use AskUserQuestion:
question: "QA Strategy Review"
options:
- "Looks good — proceed to test plan"
- "Adjust story types before proceeding"
- "Skip blocked stories and proceed with the rest"
- "Smoke check failed — fix issues and re-run /team-qa"
- "Cancel — resolve blockers first"
If smoke check FAIL: do not proceed to Phase 3. Surface the failures from the smoke check report and stop. The user must fix them, re-run /smoke-check sprint, and then re-run /team-qa.
If smoke check UNKNOWN: surface a warning — "No smoke check report found. Recommend running /smoke-check sprint before QA. Proceeding with caution."
If smoke check PASS WITH WARNINGS: note the warnings for the sign-off report and continue.
If blockers are present: list them explicitly. The user may choose to skip blocked stories or cancel the cycle.
Phase 3: Test Plan Generation
Using the strategy from Phase 2, produce a structured test plan document.
The test plan should cover:
- Scope: sprint/feature name, story count, dates
- Story Classification Table: from Phase 2 strategy
- Automated Test Requirements: which stories need test files, expected paths in
tests/ - Manual QA Scope: which stories need manual walkthrough and what to validate
- Out of Scope: what is explicitly not being tested this cycle and why
- Entry Criteria: what must be true before QA can begin. Always include: (1) Smoke check PASS or PASS WITH WARNINGS report exists at
production/qa/smoke-*.md, (2) build is stable (no crashes on launch), (3) all Must Have stories have Status: in-progress or done inproduction/sprint-status.yaml. Add any sprint-specific criteria beyond these. - Exit Criteria: what constitutes a completed QA cycle (all stories PASS or FAIL with bugs filed)
Ask: "May I write the QA plan to production/qa/qa-plan-[sprint]-[date].md?"
Write only after receiving approval.
Phase 4: Test Case Writing (qa-tester)
Smoke check is performed as part of Phase 2 (QA Strategy). If the smoke check returned FAIL in Phase 2, the cycle was stopped there. This phase only runs when the Phase 2 smoke check was PASS, PASS WITH WARNINGS, or UNKNOWN.
For each story requiring manual QA (Visual/Feel, UI, Integration without automated tests):
Spawn qa-tester via Agent for each story (run in parallel where possible), providing:
- The story file path
- The relevant section of the QA plan for that story
- The GDD acceptance criteria for the system being tested (if available)
- Instructions to write detailed test cases covering all acceptance criteria
- The output path:
production/qa/test-cases/[story-slug]-cases.md. Name it explicitly in the prompt, one per story.
Why the path is stated here rather than left to the orchestrator. The bounded write exception above holds only when "the path is one you named in the prompt". This is the phase that spawns agents in parallel, so it is where an unnamed destination does the most damage: each agent improvises its own, and two runs file the same artifact in two places. The phase reads correctly right up until two agents need somewhere to put their output.
Each test case set should include:
- Preconditions: game state required before testing begins
- Steps: numbered, unambiguous actions
- Expected Result: what should happen
- Actual Result: field left blank for the tester to fill in
- Pass/Fail: field left blank
Present the test cases to the user for review before execution. Group by story.
Use AskUserQuestion per story group (batched 3-4 at a time):
question: "Test cases ready for [Story Group]. Review before manual QA begins?"
options:
- "Approved — begin manual QA for these stories"
- "Revise test cases for [story name]"
- "Skip manual QA for [story name] — not ready"
Phase 5: Manual QA Execution
Walk through each story in the approved manual QA list.
Batch stories into groups of 3-4 and use AskUserQuestion for each:
question: "Manual QA — [Story Title]\n[brief description of what to test]"
options:
- "PASS — all acceptance criteria verified"
- "PASS WITH NOTES — minor issues found (describe after)"
- "FAIL — criteria not met (describe after)"
- "BLOCKED — cannot test yet (reason)"
After each FAIL result: use AskUserQuestion to collect the failure description, then spawn qa-tester via Agent to write a formal bug report in production/qa/bugs/.
After each PASS or PASS WITH NOTES on a Visual/Feel or UI story, write the
evidence artifact to production/qa/evidence/[story-slug]-evidence.md, from
.claude/docs/templates/test-evidence.md, carrying the sign-off table intact.
Save the screenshot you took while testing into the same directory and reference
it from the doc — /story-done checks for a retained image, not just the
write-up.
This is not optional bookkeeping — it is the artifact the next skill gates on.
/story-doneglobsproduction/qa/evidence/for Visual/Feel and UI stories and reads the sign-off table;/story-readiness,/test-evidence-reviewandgate-releaseread the same directory. Write anywhere else and a story can pass a full manual QA cycle here, then be told by/story-donethat no visual evidence exists. Visual/Feel and UI gates are BLOCKING by default, so that is a deadlock — QA passed, story cannot close. It is a merely confusing flag only wheretesting.strict.uior.visualhas been explicitly set tofalse.Leave the sign-off rows unchecked unless the sign-off actually happened in this session. An evidence file with pre-ticked approvals is worse than none: it converts a missing signature into a recorded one.
Bug report naming: BUG-[NNN]-[short-slug].md (increment NNN from existing bugs in the directory).
After collecting all results, summarize:
- Stories PASS: [count]
- Stories PASS WITH NOTES: [count]
- Stories FAIL: [count] — bugs filed: [IDs]
- Stories BLOCKED: [count]
Phase 6: QA Sign-Off Report
Spawn qa-lead via Agent to produce the sign-off report using all results from Phases 4–6.
The sign-off report format:
## QA Sign-Off Report: [Sprint/Feature]
**Date**: [date]
### Test Coverage Summary
| Story | Type | Auto Test | Manual QA | Result |
|-------|------|-----------|-----------|--------|
| [title] | Logic | PASS | — | PASS |
| [title] | Visual | — | PASS | PASS |
### Bugs Found
| ID | Story | Severity | Status |
|----|-------|----------|--------|
| BUG-001 | [story] | S2 | Open |
### Verdict: NOT ASSESSED / APPROVED / APPROVED WITH CONDITIONS / NOT APPROVED
**Conditions** (if any): [list what must be fixed before the build advances]
### Next Step
[guidance based on verdict]
Verdict rules:
Precondition, checked first. APPROVED and APPROVED WITH CONDITIONS both require that every story in scope produced executed evidence — a test that ran, or a manual case that was walked. If any story is BLOCKED, unexecuted, or has no evidence, the verdict is NOT ASSESSED and the other three rules are not evaluated.
- NOT ASSESSED — NO EVIDENCE: One or more stories produced no executed evidence (BLOCKED, tests not written, cases not walked, or smoke check FAIL / UNKNOWN). This is not a pass and not a fail; it means QA did not happen. Say which stories and why.
- APPROVED: All stories PASS or PASS WITH NOTES; no S1/S2 bugs open
- APPROVED WITH CONDITIONS: S3/S4 bugs open, or PASS WITH NOTES issues documented; no S1/S2 bugs
- NOT APPROVED: Any S1/S2 bugs open; or stories FAIL without documented workaround
Why the precondition exists. The three rules below it assume every story resolves to PASS or FAIL. A sprint where nothing was executed trips none of the NOT APPROVED conditions and vacuously satisfies "no S1/S2 bugs open" — because zero executed tests means zero observed failures. Read literally, and without this precondition, the rules let a completely untested build reach APPROVED — a
qa-leadreaching NOT APPROVED on intent would find the letter of the rules did not support it. A sign-off asserts verified quality; without this precondition the rules cannot tell "verified good" from "never looked".
Next step guidance by verdict:
- NOT ASSESSED: "QA did not run to completion. Produce the missing evidence — write the Logic tests, walk the manual cases, run
/smoke-check— then re-run/team-qa. Do not advance the build on this verdict." - APPROVED: "Build is ready for the next phase. Run
/gate-checkto validate advancement." - APPROVED WITH CONDITIONS: "Resolve conditions before advancing. S3/S4 bugs may be deferred to polish."
- NOT APPROVED: "Resolve S1/S2 bugs and re-run
/team-qaor targeted manual QA before advancing."
Ask: "May I write this QA sign-off report to production/qa/qa-signoff-[sprint]-[date].md?"
Write only after receiving approval.
Error Recovery Protocol
First, verify the artifact. If the return contract named a path, check the path exists before treating the phase as done — a named artifact that is not on disk is a failed phase, however fluent the response reads. An agent can burn a full phase and return a plausible preamble having written nothing, which is neither BLOCKED nor an error nor "cannot complete", so the trigger below never fires. Resume it naming the unmet contract; the context is usually still there.
If any spawned agent returns BLOCKED, errors, or cannot complete: surface it
immediately, don't proceed past a dependency it blocks, and always produce a
partial report. Full procedure: .claude/docs/error-recovery-protocol.md.
Common blockers:
- Input file missing (story not found, GDD absent) → redirect to the skill that creates it
- ADR status is Proposed → do not implement; run
/architecture-decisionfirst - Scope too large → split into two stories via
/create-stories - Conflicting instructions between ADR and story → surface the conflict, do not guess
Output
A summary covering: stories in scope, smoke check result, manual QA results, bugs filed (with IDs and severities), and the final APPROVED / APPROVED WITH CONDITIONS / NOT APPROVED verdict.
Verdict: COMPLETE — QA cycle finished. Verdict: BLOCKED — smoke check failed or critical blocker prevented cycle completion; partial report produced.
Session State Update
After the final phase completes (sign-off report written or BLOCKED verdict reached), silently append to production/session-state/active.md:
<!-- QA RUN: [date] | Sprint: [sprint identifier or "ad-hoc"] | Verdict: [PASS/FAIL/CONCERNS] | Report: production/qa/qa-[date].md -->