Compozy QA Bootstrap
Provision infrastructure and scratch evidence for one QA pass. Bootstrap does
not prove runtime behavior; downstream QA owns journeys and verdicts.
Required Inputs
- scenario-slug (optional): short lab context; defaults to
release-candidate.
- --playbook : required when called by
eng-real-scenario-qa; omitted for generic qa-execution infrastructure.
- --profile targeted --required-surface : use for a bounded non-agent journey; repeat the surface flag for every CLI/API/Web/runtime/provider plane in scope.
- Provider resources are opt-in through the scenario contract:
feature, broad, and playbook profiles require live providers; targeted labs opt in only with --required-surface provider. Other targeted labs receive no copied login or global Codex skills/cache links.
- Provider usage collection is disabled before daemon startup in every profile. Pass
--enable-provider-usage only when a live-provider scenario explicitly needs account usage collection.
Procedure
Step 1: Create or Resume One Lab
- Resolve the scenario slug and, for
eng-real-scenario-qa, validate the playbook under .agents/skills/eng/eng-real-scenario-qa/references/playbooks/.
- For a new pass, run the bootstrap helper (bootstrap, mutating):
python3 .agents/skills/eng/eng-qa-bootstrap/scripts/bootstrap-qa-env.py --scenario "<scenario-slug>" --repo-root . [--playbook "<ref>" | --profile targeted --required-surface "<surface>"]
- Reuse only when continuing the same active QA session or loop with its exact manifest:
python3 .agents/skills/eng/eng-qa-bootstrap/scripts/bootstrap-qa-env.py --scenario "<scenario-slug>" --repo-root . [--playbook "<ref>"] --reuse-manifest "<manifest-path>"
- Record the emitted
BOOTSTRAP_MANIFEST; do not reconstruct environment values manually.
Done when: one fresh or same-session lab has one readable manifest and no unrelated lab was reused.
Step 2: Verify the Handoff Contract
- Read
.agents/skills/eng/eng-qa-bootstrap/references/bootstrap-contract.md in full.
- Check that the emitted manifest, required paths, launch policy, and teardown command match the selected profile. Use helper diagnostics and the strict auditor for machine-checkable fields; do not manually repeat each schema check.
- When a playbook was supplied, also validate the materialized playbook, agents, open-task tree, required deliverables/collaboration, and populated charter.
- Treat scaffolding as empty evidence until downstream QA records real actions and the strict auditor passes.
Done when: manifest, env, paths, scenario contract, charter, playbook artifacts, and filesystem all describe the same isolated lab with no placeholders in a playbook run.
Step 3: Hand Off to Downstream QA
- Pass
qa-docs-path=docs/qa to qa-report and qa-execution; keep QA_OUTPUT_PATH as lab-side scratch evidence only.
- Launch providers, Web, browser, and config operations exactly from the manifest contract. Preserve operator home for
native_cli + home_policy=operator; use isolated provider homes for bound-secret or brokered lanes.
- Export the manifest-derived
COMPOZY_WEB_API_PROXY_TARGET for Web QA and serialize config writes against one isolated home.
- Register every long-lived process immediately under
<QA_OUTPUT_PATH>/qa/pids/.
- Require downstream QA to run the manifest's strict
AUDIT_COMMAND before a behavior-first verdict.
Done when: downstream skills consume the canonical manifest, durable findings land under docs/qa/, scratch evidence stays in the lab, and every owned process is registered.
Step 4: Preserve Continuation State
- Report the manifest path, lab root,
COMPOZY_HOME, base URL, dated run-report path, and reuse status.
- For a continuing loop, append the exact machine-readable continuation block from the bootstrap contract. Hand off with the lab alive; Step 5 runs when downstream QA reaches a terminal outcome, not before its first action.
Done when: the next continuation can identify the same lab from one exact manifest without discovery or guessing.
Step 5: Tear Down on Every Terminal Path
- On PASS, FAIL, BLOCKED, or abort, run the manifest's exact
TEARDOWN_COMMAND; only a confirmed continuation of the same loop may defer it.
- The teardown must stop the daemon, tmux server, registered PIDs, lab-root processes, and lab-port listeners without touching unrelated labs.
- Cite
<QA_OUTPUT_PATH>/qa/teardown.json with "clean": true. Teardown removes copied provider files and global skills/cache links; process survivors or provider resource removal errors are blocking failures.
- Reserve
make qa-reap for intentional stale-lab recovery across the machine, not normal cleanup of one active run.
Done when: the current lab's teardown evidence is clean and no process owned by this pass remains alive.
Process identity
Register every owned process immediately with
python3 .agents/skills/eng/eng-qa-bootstrap/scripts/qa_process_identity.py --pid "$!" --file "<QA_OUTPUT_PATH>/qa/pids/<name>.pid".
The record includes start time and command. Teardown revalidates it before each individual signal,
never signals whole process groups, and reports matched, stale, and skipped records. Legacy bare
PID files and port listeners require independent lab ownership evidence. Dead process records are
removed during teardown; do not reuse a record for a replacement process.
Error Handling
- Requested reuse fails health checks: use the fresh manifest emitted by the helper; do not revive stale state manually.
- Provider reports global config errors: compare the command's home policy with the manifest; isolated and operator-home lanes intentionally differ.
- Web reaches the wrong daemon: restart the Web process with the manifest-derived proxy target.
- Teardown reports survivors: inspect each
TEARDOWN_SURVIVOR, stop only processes owned by the current lab, and rerun the same targeted teardown.
1---2name: eng-qa-bootstrap3description: Bootstraps isolated Compozy QA labs and emits the canonical manifest for downstream QA. Use when local QA needs daemon, workspace, provider, or playbook setup. Do not use for unit tests, planning-only work, or browser-only checks.4trigger: explicit5argument-hint: "[scenario-slug] [--playbook <ref> | --profile targeted --required-surface <surface>]"6metadata:7 source: https://skillmd.com/skills/compozy/eng-qa-bootstrap8---910# Compozy QA Bootstrap1112Provision infrastructure and scratch evidence for one QA pass. Bootstrap does13not prove runtime behavior; downstream QA owns journeys and verdicts.1415## Required Inputs1617- **scenario-slug** (optional): short lab context; defaults to `release-candidate`.18- **--playbook <ref>**: required when called by `eng-real-scenario-qa`; omitted for generic `qa-execution` infrastructure.19- **--profile targeted --required-surface <surface>**: use for a bounded non-agent journey; repeat the surface flag for every CLI/API/Web/runtime/provider plane in scope.20- Provider resources are opt-in through the scenario contract: `feature`, `broad`, and playbook profiles require live providers; targeted labs opt in only with `--required-surface provider`. Other targeted labs receive no copied login or global Codex skills/cache links.21- Provider usage collection is disabled before daemon startup in every profile. Pass `--enable-provider-usage` only when a live-provider scenario explicitly needs account usage collection.2223## Procedure2425**Step 1: Create or Resume One Lab**26271. Resolve the scenario slug and, for `eng-real-scenario-qa`, validate the playbook under `.agents/skills/eng/eng-real-scenario-qa/references/playbooks/`.282. For a new pass, run the bootstrap helper (bootstrap, mutating):29 `python3 .agents/skills/eng/eng-qa-bootstrap/scripts/bootstrap-qa-env.py --scenario "<scenario-slug>" --repo-root . [--playbook "<ref>" | --profile targeted --required-surface "<surface>"]`303. Reuse only when continuing the same active QA session or loop with its exact manifest:31 `python3 .agents/skills/eng/eng-qa-bootstrap/scripts/bootstrap-qa-env.py --scenario "<scenario-slug>" --repo-root . [--playbook "<ref>"] --reuse-manifest "<manifest-path>"`324. Record the emitted `BOOTSTRAP_MANIFEST`; do not reconstruct environment values manually.3334*Done when:* one fresh or same-session lab has one readable manifest and no unrelated lab was reused.3536**Step 2: Verify the Handoff Contract**37381. Read `.agents/skills/eng/eng-qa-bootstrap/references/bootstrap-contract.md` in full.392. Check that the emitted manifest, required paths, launch policy, and teardown command match the selected profile. Use helper diagnostics and the strict auditor for machine-checkable fields; do not manually repeat each schema check.403. When a playbook was supplied, also validate the materialized playbook, agents, open-task tree, required deliverables/collaboration, and populated charter.414. Treat scaffolding as empty evidence until downstream QA records real actions and the strict auditor passes.4243*Done when:* manifest, env, paths, scenario contract, charter, playbook artifacts, and filesystem all describe the same isolated lab with no placeholders in a playbook run.4445**Step 3: Hand Off to Downstream QA**46471. Pass `qa-docs-path=docs/qa` to `qa-report` and `qa-execution`; keep `QA_OUTPUT_PATH` as lab-side scratch evidence only.482. Launch providers, Web, browser, and config operations exactly from the manifest contract. Preserve operator home for `native_cli + home_policy=operator`; use isolated provider homes for bound-secret or brokered lanes.493. Export the manifest-derived `COMPOZY_WEB_API_PROXY_TARGET` for Web QA and serialize config writes against one isolated home.504. Register every long-lived process immediately under `<QA_OUTPUT_PATH>/qa/pids/`.515. Require downstream QA to run the manifest's strict `AUDIT_COMMAND` before a behavior-first verdict.5253*Done when:* downstream skills consume the canonical manifest, durable findings land under `docs/qa/`, scratch evidence stays in the lab, and every owned process is registered.5455**Step 4: Preserve Continuation State**56571. Report the manifest path, lab root, `COMPOZY_HOME`, base URL, dated run-report path, and reuse status.582. For a continuing loop, append the exact machine-readable continuation block from the bootstrap contract. Hand off with the lab alive; Step 5 runs when downstream QA reaches a terminal outcome, not before its first action.5960*Done when:* the next continuation can identify the same lab from one exact manifest without discovery or guessing.6162**Step 5: Tear Down on Every Terminal Path**63641. On PASS, FAIL, BLOCKED, or abort, run the manifest's exact `TEARDOWN_COMMAND`; only a confirmed continuation of the same loop may defer it.652. The teardown must stop the daemon, tmux server, registered PIDs, lab-root processes, and lab-port listeners without touching unrelated labs.663. Cite `<QA_OUTPUT_PATH>/qa/teardown.json` with `"clean": true`. Teardown removes copied provider files and global skills/cache links; process survivors or provider resource removal errors are blocking failures.674. Reserve `make qa-reap` for intentional stale-lab recovery across the machine, not normal cleanup of one active run.6869*Done when:* the current lab's teardown evidence is clean and no process owned by this pass remains alive.7071## Process identity7273Register every owned process immediately with74`python3 .agents/skills/eng/eng-qa-bootstrap/scripts/qa_process_identity.py --pid "$!" --file "<QA_OUTPUT_PATH>/qa/pids/<name>.pid"`.75The record includes start time and command. Teardown revalidates it before each individual signal,76never signals whole process groups, and reports matched, stale, and skipped records. Legacy bare77PID files and port listeners require independent lab ownership evidence. Dead process records are78removed during teardown; do not reuse a record for a replacement process.7980## Error Handling8182- **Requested reuse fails health checks:** use the fresh manifest emitted by the helper; do not revive stale state manually.83- **Provider reports global config errors:** compare the command's home policy with the manifest; isolated and operator-home lanes intentionally differ.84- **Web reaches the wrong daemon:** restart the Web process with the manifest-derived proxy target.85- **Teardown reports survivors:** inspect each `TEARDOWN_SURVIVOR`, stop only processes owned by the current lab, and rerun the same targeted teardown.