refactor(daemon-contracts): drop the registration and claims fixture exports

[thymikee]

Stacked on #3686 (test-kit stops depending on daemon-contracts, which this PR needs: daemon-contracts now declares test-kit as a devDependency, and test-kit → daemon-contracts would be a project-reference cycle). Base is #3686's branch; GitHub retargets to main when it merges. Review only the last four commits.

What

Removes daemon-contracts/daemon-registration-fixtures and daemon-contracts/device-claims-fixtures (daemon-contracts exports 17 → 15), moves the one test of daemon-contracts' own code out of daemon, and replaces the package's duplicated tmp-dir.fixtures.ts with test-kit's.

Fixture decisions

Fixture	Importers by package	Outcome	Reason
publishDaemonRegistration	daemon-contracts 1 (own test), daemon 5, root 1 (cli-device-status.test.ts)	test-kit named export	Shared across packages; daemon-contracts' own test now uses the same one, so the roughly 15-line duplicate the coordinator accepted is not needed.
resolveDeviceClaimPath	daemon 7 (6 tests and session-idle-expiry-harness)	test-kit deviceClaimFilePath	Shared across daemon tests; restates the sha256 layout and AGENT_DEVICE_CLAIMS_DIR override so tests pin the on-disk contract instead of re-importing the production resolver. daemon-contracts' own tests never used it.
createDeviceClaimReconciler	1 daemon test	Case 2, moved into daemon-contracts	The test exercises only device-claim-reconciliation.ts. The coordinator chose to rewrite it through device-claim-recovery because daemon-contracts could not import test-kit; with #3686 it can, so the test moves as a pure git mv instead (named for its module) and keeps testing the unit directly.

Commits

1. git mv of the reconciler test into packages/daemon-contracts/src/device-claim-reconciliation.test.ts (pure, 100%).
2. test-kit gains publishDaemonRegistration and deviceClaimFilePath; daemon and root importers switch to @agent-device/test-kit (specifiers, plus the resolveDeviceClaimPath → deviceClaimFilePath identifier at its call sites); the moved test's two specifiers change (./device-claim-reconciliation.ts, ./device-claims.ts); daemon-contracts declares test-kit (devDependencies + tsconfig reference + lockfile).
3. chore(gates): delete daemon-registration.fixtures.ts, device-claims.fixtures.ts and their two exports.
4. Delete tmp-dir.fixtures.ts; six daemon-contracts tests import mkdtempForTestSync from test-kit.

Test-count parity (vitest run … --reporter=json)

All files this diff touches, plus every daemon-contracts test: 194 passed / 0 failed on origin/main (23 files, with the reconciler test at its daemon path), 194 / 0 here (same files, reconciler test at its new path; plus daemon-registration-owner.test.ts's 47, covered by #3686). Reconciler test alone: 2 / 0 before and after.

Selection: packages/daemon-contracts/src/device-claim-reconciliation.test.ts is under packages/*/src/**/*.test.ts (unit-core), collected by path; check:affected --run includes it. vi.mock: the moved test uses vi.fn only; none of the touched files adds one.

Surface

daemon-contracts exports 17 → 15; no new subpath on any package. test-kit adds two named exports from its one entry (sorted position). No barrels or export *. daemon-contracts: devDependencies and tsconfig references both declare test-kit.

Gates (on ff0920312, rebased on main after #3684)

check:quick, check:layering, check:fallow, check:gate-manifest, pnpm format, check:affected --run (exit 0; 1828 files, 15474 tests): all clean.

🤖 Generated with Claude Code

[cubic-dev-ai]

No issues found across 28 files

_{View guided diff | Turn on auto-fix | Re-trigger cubic}

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-16 06:04 UTC

[thymikee]

This PR is ready at ff09203. I found no problems in the diff.

Not blocking: deviceClaimFilePath in packages/test-kit/src/device-claim-fixtures.ts copies the path logic of resolveDeviceClaimRoot and resolveDeviceClaimPath in daemon-contracts. That looks deliberate, since it gives tests an independent check of the on-disk layout and keeps test-kit free of the daemon-contracts import that #3686 removed. If you want drift to fail in one place, a daemon-contracts test could assert that deviceClaimFilePath(k) equals resolveDeviceClaimPath(k). You can take this or leave it.

CI is still running on this head: 14 of 18 checks are queued or in progress, and none have failed. I did not run the typecheck, layering, fallow or unit gates myself. The 194/194 parity numbers come from the PR body and I did not reproduce them. This PR is stacked on #3686, and I only checked that test-kit has no daemon-contracts dependency at this head. I did not review #3686's own diff. No conflicts. Before merge, CI needs to finish green on this head and #3686 needs to land, after which GitHub will retarget this PR to main.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.24 MB	5.24 MB	0 B
Package (unpacked)	5.24 MB	5.24 MB	0 B
Package (download)	1.57 MB	1.57 MB	+13 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	29.4 ms	29.7 ms	+0.3 ms
CLI --help	82.4 ms	82.0 ms	-0.4 ms

[thymikee]

Took the drift-test suggestion: packages/daemon-contracts/src/device-claim-paths.test.ts (named for device-claim-paths.ts) asserts deviceClaimFilePath(k) equals resolveDeviceClaimPath(k), with and without AGENT_DEVICE_CLAIMS_DIR set. Mutation check: switching test-kit's hash from sha256 to sha1 fails both cases; reverted. 2/2 pass, plus the earlier 194/194 parity set.

On the new head check:quick, check:layering, check:fallow, check:gate-manifest and pnpm format are clean. Two full check:affected --run passes failed only with 5s/30s timeouts in unrelated suites (iOS lifecycle, fuzz harness, adb host, claim-recovery composition) while the host load average was ~76; a different set failed each time, and every one passes in isolation (38/38 across those five files). CI is the authority on this head.

[thymikee]

The code looks good at 701802c. This follows my earlier review at ff09203, and nothing from that review is left open.

All 23 checks pass. I did not run the drift test or the sha1 mutation locally, so I relied on your report and on green CI. I also did not run the layering scan locally, so I relied on CI and a grep.

There are no conflicts. #3689 is stacked on #3686, so it can merge once that base lands.