fix(test): correct gh-aw compile test workspace layout for dispatch-workflow validation

[tamirdresher]

Root cause

test/gh-aw-quality.test.ts -> gh-aw: compiled workflow contract -> strict-compiles and preserves prompt/config behavior has kept dev red on vitest since 2026-08-12T23:39 (PR CI, not the node --test release gate, which does not run this file - see note below).

The test builds an ephemeral git workspace, copies this repo's workflows/ into <workspace>/workflows/, then runs gh aw compile workflows/squad.md --strict.

#1682 added safe-outputs.dispatch-workflow: { workflows: [squad-implement-worker] } to workflows/squad.md. gh aw compile --strict (local version v0.85.4) validates that dispatch target against a .github/workflows/ directory it computes relative to the compiled file, assuming the standard depth <root>/.github/workflows/<file>.md. This repo's real gh-aw source lives one level shallower, at <root>/workflows/<file>.md, so gh aw's computed root lands one directory too high - the parent of the actual workspace/repo root. It can never find squad-implement-worker, and --strict compilation fails:

dispatch-workflow: workflow 'squad-implement-worker' not found in <...>/.github/workflows
Checked for: squad-implement-worker.md, .lock.yml, .yml, .yaml

I reproduced this locally in a clean temp git repo containing only workflows/, and independently confirmed the off-by-one behavior against the real repo root as well. squad-implement-worker.md is present and committed on dev - this was never a missing-source problem, only a test-workspace-layout problem.

What I changed (and why)

Only test/gh-aw-quality.test.ts. workflows/squad.md, workflows/squad-implement-worker.md, and #1682's dispatch-workflow feature surface are completely untouched - I did not revert, weaken, or touch the feature in any way.

In the compile test:

• Copy workflows/ into <workspace>/.github/workflows/ instead of <workspace>/workflows/.
• Compile .github/workflows/squad.md (was workflows/squad.md), and read the resulting lock file from the matching path.
• This mirrors exactly how real downstream consumers install and compile this repo's workflows per docs/src/content/docs/guide/gh-aw.md: gh aw add owner/squad/workflows/squad-implement-worker.md@dev owner/squad/workflows/squad.md@dev lands both .md files together in the consumer's own .github/workflows/, alongside shared/*.md, which is the layout gh aw compile --strict's dispatch-workflow validation actually expects.
• I verified this placement in a fully isolated compile (outside this repo/worktree entirely): with .github/workflows/ layout, gh aw compile .github/workflows/squad.md --strict succeeds (exit 0, 1 succeeded, 2 warnings, dispatch-workflow resolves).
• That same isolated compile also revealed a second, previously-passing assertion in this test needed updating: gh aw records {{#runtime-import ...}} placeholder paths relative to the repo root, not relative to the compiled file's directory. With this real layout the compiled output legitimately contains {{#runtime-import .github/workflows/shared/planning-ontology.md}} and {{#runtime-import .github/workflows/squad.md}} (previously asserted as the bare shared/planning-ontology.md / squad.md, which only held for the old, incorrect bare-workflows/ test layout). Updated both literal assertions to match gh aw's genuine, verified behavior.
• Added a 20s explicit test timeout (gh aw compile subprocess reliably takes ~2-3s, but can exceed vitest's 5s default under full-suite parallel resource contention on some machines - observed this directly while re-running the full suite for verification; this is a low-risk hardening of the exact test being touched, not scope creep).

Local verification

• npx vitest run test/gh-aw-quality.test.ts -t "strict-compiles": 1 passed, 78 skipped.
• Full gh-aw-quality.test.ts file: the only remaining failures (12) are pre-existing, unrelated, Windows-only execSync(..., { shell: '/bin/sh' }) calls (Team Guard roster-row detection, Cast PR dedup jq filter) that ENOENT because /bin/sh doesn't exist on Windows. These are not caused by this change and were already confirmed (via the real failing CI run's logs) not to occur on Linux CI, where only this one dispatch-workflow test failed.
• Full npx vitest run (all ~272 test files, ~7500 tests) on this Windows dev machine: pre-existing, unrelated flakiness exists independent of this change (roughly 60-65 of 272 files fail even on a clean dev checkout, mostly Windows//bin/sh and environment-specific issues) - I confirmed this is not a regression by diffing my change against a stashed/reverted state and re-running. My diff is a single, 26-line change to one test's workspace setup and two literal assertions; nothing else was touched.
• Note for reviewers: running the full suite also has a pre-existing side effect where some build/test step rewrites ~150 template files in the working tree with different line endings, and stamps a version comment into .github/agents/squad.agent.md. This is unrelated repo behavior (not part of my diff) - I reverted it every time it appeared and confirmed via git diff --stat -w that it is pure line-ending churn plus one unrelated version-stamp side effect, before committing.

Note on why dev looked green on the release gate despite this

squad-release.yml runs node --test test/*.test.cjs, while PR CI runs vitest. test/gh-aw-quality.test.ts is a .ts file the release gate's glob never sees, so this red vitest test did not block dev -> main promotion (e.g. the v0.12.0 release-prep PR #1691 merged despite it). Worth knowing if dev looks green from the release workflow's perspective while PR CI disagrees.

Confirming #1682 is untouched

• git diff for this PR touches only test/gh-aw-quality.test.ts.
• workflows/squad.md's dispatch-workflow, issue_number/aw_context inputs, and bots: trigger are all present, unmodified, on this branch.
• workflows/squad-implement-worker.md is unmodified.

@bradygaster - flagging directly since you own #1682 and the dispatch-workflow feature; this PR only adjusts how the test verifies it compiles, it does not change the feature's behavior or configuration.

[github-actions]

🟢 Impact Analysis — PR #1697

Risk tier: 🟢 LOW

📊 Summary

Metric	Count
Files changed	1
Files added	0
Files modified	1
Files deleted	0
Modules touched	1

🎯 Risk Factors

• 1 files changed (≤5 → LOW)
• 1 module(s) touched (≤1 → LOW)

📦 Modules Affected

tests (1 file)

• test/gh-aw-quality.test.ts

---

This report is generated automatically for every PR. See #733 for details.

[github-actions]

🛫 PR Readiness Check

ℹ️ This comment updates on each push. Last checked: commit 2ccdd37

PR Scope: 🔧 Infrastructure

⚠️ 2 item(s) to address before review

Status	Check	Details
✅	Single commit	1 commit — clean history
✅	Not in draft	Ready for review
✅	Branch up to date	Up to date with dev
❌	Copilot review	No Copilot review yet — it may still be processing
✅	Changeset present	No source files changed — changeset not required
✅	Scope clean	No .squad/ or docs/proposals/ files
✅	No merge conflicts	No merge conflicts
❌	Copilot threads resolved	1 unresolved Copilot thread(s) — fix and resolve before merging
✅	CI passing	All checks passing

Files Changed (1 file, +26 −6)

File	+/−
test/gh-aw-quality.test.ts	+26 −6

Total: +26 −6

---

This check runs automatically on every push. Fix any ❌ items and push again.
See CONTRIBUTING.md and PR Requirements for details.

[Copilot]

Pull request overview

Adjusts the gh aw compile --strict contract test workspace layout to mirror real downstream installs (.github/workflows/), so dispatch-workflow validation can resolve squad-implement-worker correctly and the test stops failing on CI.

Changes:

• Copy workflows/ into <workspace>/.github/workflows/ and compile .github/workflows/squad.md in the ephemeral test repo.
• Update runtime-import literal expectations to match paths recorded relative to the repo root under the new layout.
• Add an explicit 20s timeout to reduce flakiness from gh aw compile subprocess duration under contention.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.