osovv / grace-verification

<skill>
<purpose>
Strengthen deterministic verification for modules and changes. Verification state lives in .grace/verification/index.xml and routed verification documents. Each durable module should have deterministic V-M-* coverage unless an explicit exception is planned.
</purpose>

<workflow>

1. Read relevant .grace/graph anchors and current V-M-* entries.
2. Identify scenarios, commands, test files, required log markers, and trace assertions.
3. Ensure commands are deterministic and runnable from the project root or documented cwd.
4. Update or propose .grace/verification changes through the active change plan.
5. Run the commands and record fresh evidence in the response.

</workflow>
<cwd_contract>
When verification commands run from a workspace or package directory, add one direct <Cwd>relative/project/path</Cwd> child to the owning V-M-* entry. Keep declared <TestFiles><File>...</File></TestFiles> paths project-root-relative; the CLI uses Cwd only to compare them with cwd-relative command arguments.
</cwd_contract>
<evidence_contract>
Use <Marker> when module health must prove a runtime log or trace emission from linked implementation code. Use <TraceAssertion> for deterministic test or trace evidence that does not require runtime logging, such as pure functions, type-level modules, and core libraries. A non-empty marker or trace assertion satisfies the module-health evidence requirement; only authored markers require matching runtime emission and BLOCK_* evidence.
</evidence_contract>
<gatetaskcontract>
Reference named project tasks in MustPassCommand and ExpectedCommand (for example bun run gate:e2e) instead of inline ;-chains. Decompose full gates into granular named tasks (gate:test, gate:typecheck, gate:build, gate:e2e) so a selective re-run is just running that task, timeouts map to one coherent unit, and grace lint --run-commands reports each gate step with its own timing and log.
</gatetaskcontract>
<flake_contract>
A flaky gate is a defect of the verification entry, not a reason to re-run blindly. Diagnose from the per-attempt logs under ~/.cache/grace/run-commands/. Fix it by decomposing the gate task, adding runner-level retries inside the task itself (for example playwright --retries), or quarantining the unstable scenario. Silent manual re-runs are not a fix: grace records every run honestly, and nondeterministic verification undermines assertion evidence.
</flake_contract>
</skill>