Give the Windows supervisor job room to finish

[BunsDev]

Closes #324.

Windows supervisor behavior is capped at timeout-minutes: 20 and needs 18 to 20, so it dies on the cap. GitHub reports the overrun as cancelled, which reads as a flake. It has taken out #313 three times and main twice, always after every assertion passed, in the final cargo check step.

Raises the cap to 30 on that job only.

Two commits, as the protocol requires

.github/workflows/ci.yml is one of the 25 files under harnessAuthority, so this follows the two-step in docs/phase1-conformance.md:

• 74689b1 — the edit, plus ci.yml's new blob and SHA-256 in the lock and in the test's literal copy. Every digest rides with the change itself.
• ef7bd74 — repin only: harness.revision, harnessAuthority.revision/.tree, and the prose and test copies of them. No digests, no behaviour.

Verification

Check	Result
All 25 governed digests vs authority commit 74689b1	all match
Recorded tree vs 74689b1^{tree}	identical
vitest --config vitest.heavy.config.ts src/phase1-conformance-lock.test.ts	95 passed, 1 skipped
pnpm exec vitest run	1321 passed, 63 skipped
pnpm typecheck, pnpm lint	clean

There is a second timeout-minutes: 20 at line 129 of the same file, so the edit was made by line number rather than string match; the diff touches one line. No test pinned this job's timeout.

Merge requirement

The repin points harnessAuthority.revision at 74689b1, a commit on this branch. Per docs/phase1-conformance.md it must land as a real merge commit, never a squash or rebase, or the pinned revision stops being an ancestor of main and the authority checkout only resolves while this branch survives. An SDK rebind follows, as it does for any authority move.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The workflow timeout adjustment is narrowly scoped and the governed conformance lock/test/docs repins are updated consistently with the change.

Pull request overview

This PR increases the job-level timeout for the “Windows supervisor behavior” GitHub Actions job to prevent it from being cancelled at the 20-minute cap, and then updates the Phase 1 conformance authority pinning artifacts to keep governed workflow digests and documented bindings consistent.

Changes:

• Raise .github/workflows/ci.yml job windows-supervisor-behavior timeout-minutes from 20 → 30.
• Repin Phase 1 conformance harness authority (revision/tree) and update the governed blob/SHA-256 entry for the changed workflow in phase1-conformance.lock.json and its test mirror.
• Refresh the documented “reviewed source” binding in docs/phase1-conformance.md to match the new authority revision/tree.

File summaries

File	Description
src/phase1-conformance-lock.test.ts	Updates the expected harness authority revision/tree and the pinned blob/SHA-256 for .github/workflows/ci.yml used by the lock tests.
phase1-conformance.lock.json	Repins harness.revision and harnessAuthority (revision/tree), plus the .github/workflows/ci.yml blob/SHA-256 entry to reflect the workflow change.
docs/phase1-conformance.md	Updates the prose binding to the new reviewed source revision/tree.
.github/workflows/ci.yml	Increases windows-supervisor-behavior job timeout to 30 minutes to avoid end-of-job cancellation.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 0
• Review effort level: Lite

---

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

[BunsDev]

Force-pushed a correction. The first attempt failed Web checks in the heavy suite with Chat conformance native delta does not match its immutable allowlist, and the cause is worth recording.

harnessAuthority.productionDeltas carries ten digests alongside the 25 governed files. Three of them were stale: src-tauri/Cargo.toml and Cargo.lock from 07ffd59, and src-tauri/src/keyring.rs from 9761e87, all landed in #322 without a repin. That stayed invisible because the authority still pointed at b5e0fac, so the deltas were verified against that older tree and matched it correctly. Advancing the authority to a commit on current main is what surfaced the drift.

Their digests now move in the code commit, where digests belong, rather than in the repin. The repin still touches only the revision, the tree, and their prose and test copies.

Rebuilt rather than patched on top, so the two-step shape is preserved:

• a3dc4d9 — the edit, ci.yml's digests, and the three refreshed deltas
• 285ced0 — repin only

Re-verified against the new authority: tree matches a3dc4d9^{tree}, and all 25 governed files plus all 10 production deltas match. Heavy suite 909 passed, typecheck and lint clean.

Two tests, windows-shared-quota and windows-process-termination-race, fail only when the normal and heavy suites run back to back locally and pass individually. They are untouched by this branch.