docs(release-process): add post-fix repro verification policy

Race-condition and concurrency fixes must include reporter-verified repro
confirmation before the originating issue is closed. CI green is necessary
but not sufficient.

Closes M12

docs/reference/release-process.md

@@ -1,79 +1,44 @@
-# Release Process Reference
+# Release Process
 
-This reference captures release gates that are easy to miss during urgent fixes.
+This reference records release gates that are not covered by CI alone.
+
+## Standard Release Gates
+
+Before publishing a release, maintainers verify:
+
+- Version bump and package metadata are present on the release branch.
+- Targeted tests for changed code pass.
+- `bun run typecheck` passes.
+- User-facing documentation covers new public behavior.
+- Known issues are documented before the release notes are finalized.
+
+CI green is required for release readiness, but CI does not replace manual
+verification for bugs whose reproducer depends on timing, providers, models, or
+external OpenCode behavior.
 
 ## Post-Fix Repro Verification
 
-For race-condition and concurrency fixes, CI green is necessary but not
+### Policy
-sufficient.
 
-Before closing the source issue, the original issue reporter must re-run the
+For race-condition and concurrency fixes, the original issue reporter, or a
-documented reproducer against the fix commit. If the reporter is unavailable, a
+maintainer if the reporter is unavailable, must re-run the documented
-maintainer must run the same reproducer in an equivalent environment.
+reproducer against the fix commit before the issue is closed. CI green is
+necessary but not sufficient.
 
-This policy applies to bugs involving:
+### Checklist
 
-- duplicate streaming output
+- [ ] Reproducer documented in the issue thread with steps, expected result,
-- repeated internal prompt injection
+  and actual result.
-- session recovery races
+- [ ] Fix commit identified.
-- background task wake races
+- [ ] Reproducer re-run on the fix commit.
-- runtime fallback retry races
+- [ ] Result documented in the issue thread as
-- team mailbox delivery races
+  "Repro retested: PASS on <commit-sha>".
-- test contamination caused by shared mocks or module state
+- [ ] If the repro is environmental, such as a specific OS, model, or provider,
+  the re-run is attempted in matching conditions.
 
-### Required checklist
+### Escalation
 
-- Record the issue number and fix commit hash.
+If the repro cannot be obtained, such as a transient race that does not
-- Confirm the reproducer is documented in the issue or PR.
+reproduce locally, the limitation must be noted in the issue close comment and
-- Build or install the exact fix commit under test.
+added to release notes as "Fix unverified end-to-end". Do not close the issue
-- Run the reproducer without local patches.
+as fully verified without that disclosure.
-- Capture the command, input prompt, config, provider, model, and platform.
-- Confirm the original failure is absent.
-- Confirm no new adjacent failure appears in logs or terminal output.
-- Link the successful repro result before closing the issue.
-
-### Reporter path
-
-1. Ask the original reporter to test the fix commit.
-2. Provide exact install or checkout instructions.
-3. Ask for terminal output, logs, or a short screen recording when relevant.
-4. Close the issue only after the reporter confirms the failure no longer
-   reproduces.
-
-### Maintainer fallback path
-
-Use this path when the reporter is unavailable or cannot test the fix.
-
-1. Recreate the reported environment as closely as practical.
-2. Use the same provider and model class if provider behavior is part of the
-   failure.
-3. Run the documented reproducer against the fix commit.
-4. Attach the maintainer repro notes to the issue.
-5. State which parts of the environment could not be matched.
-
-### Environmental escalation path
-
-If the reproducer depends on unavailable local state, credentials, provider
-behavior, timing, or platform details:
-
-1. Keep the issue open.
-2. Label or note it as environment-dependent repro pending.
-3. Ask for the missing environment details or sanitized logs.
-4. Add a maintainer-owned minimized reproducer if one can be derived.
-5. Land extra diagnostics when the failure cannot be observed directly.
-6. Re-run the repro after diagnostics or environment access is available.
-
-Do not close a race-condition issue based only on unit tests, typecheck, or a
-green CI run.
-
-### Closure note template
-
-```text
-Post-fix repro verification:
-- Issue: #<number>
-- Fix commit: <hash>
-- Verified by: <reporter or maintainer>
-- Environment: <os, runtime, provider, model>
-- Reproducer: <link or summary>
-- Result: original failure no longer reproduces
-```