# v0.3.9-alpha.7 Release Notes

## Version

`v0.3.9-alpha.7`

## Dist Tag

`alpha`

During the alpha pilot, `alpha` is the tester-facing dist-tag. `latest` may lag
behind `alpha`, so verify both before treating default install as current.

## Summary

This prerelease packages the read-only workflow closure planner added after the
alpha6 single implementation-to-finish trial showed that checklist wording alone
was not enough to bring a model back to workflow closure.

Included commits:

- `5df7dc1 feat(cli): add workflow closure planner`;
- `b3f24c7 fix(cli): stabilize workflow closure JSON schema`;
- `d5e5150 docs: record workflow closure planner contract`.

Persona Harness remains an AI coding workflow rail + evidence + continuation
harness. This release is not eval proof, PH superiority proof, generated app
product-quality certification, AST/linter enforcement, or an OpenCode/model
quality claim.

## What Changed

### Read-only closure planner

- `workflow closure status --json` exposes the current closure state and
  blockers.
- `workflow closure next --json` exposes deterministic next steps and the first
  actionable blocker.
- JSON state keys are stable, including `currentTicket: null` when no current
  ticket exists.
- Verification remains evidence-backed: command names alone are not treated as
  PASS evidence.
- `workflow finish implement` remains the final authority.

### Boundaries

- No `closure run` command is included.
- No reports are auto-filled.
- No tickets are auto-archived.
- No workflow is auto-finished.
- No finish gate is weakened.
- The alpha6 single finish trial remains fail-to-finish. Alpha7 is the
  structural follow-up, not retroactive success for that trial.

## Verification

QA re-verification before release prep passed for:

- closure JSON smoke;
- `npm run smoke:product-mvp`;
- focused workflow closure tests;
- `npm run typecheck`.

Release-prep verification for this worktree:

```bash
npm test
npm run typecheck
npm run build
npm run check:docs
npm run smoke:product-mvp
npm pack --dry-run
git diff --check
```

Observed result on 2026-06-29:

- `npm test`: passed, 70 files / 463 tests.
- `npm run typecheck`: passed.
- `npm run build`: passed.
- `npm run check:docs`: passed.
- `npm run smoke:product-mvp`: passed.
- `npm pack --dry-run`: passed for `persona-harness@0.3.9-alpha.7`.
  - filename: `persona-harness-0.3.9-alpha.7.tgz`;
  - package size: `387.8 kB`;
  - unpacked size: `1.6 MB`;
  - shasum: `77baf2792ab539d3e20b7799c7fe1ce291945ab6`;
  - total files: `377`.
- `git diff --check`: passed.

## Publish Notes

- Publish with `npm publish --tag alpha` only after all verification commands
  pass.
- Do not intentionally move `latest` during this manual release unless
  separately approved.
- If npm requires OTP/2FA, stop after the first failure and hand off the exact
  command for the user to run.

## Published Registry Facts

`persona-harness@0.3.9-alpha.7` exists on the npm registry.

Registry facts observed after user-completed publish:

- version: `0.3.9-alpha.7`;
- gitHead: `56f38eb2f37d217267cb51ee347538a18114625f`;
- dist.shasum: `85af4864892a938310c0f1f87ead7bca8cf7a9cb`;
- dist-tags: `alpha=0.3.9-alpha.7`, `latest=0.3.9-alpha.3`.

Git facts:

- local tag `v0.3.9-alpha.7` points at
  `56f38eb2f37d217267cb51ee347538a18114625f`;
- origin tag `refs/tags/v0.3.9-alpha.7` points at
  `56f38eb2f37d217267cb51ee347538a18114625f`;
- origin/main includes the release prep commit
  `56f38eb2f37d217267cb51ee347538a18114625f`.

`latest` was not moved. This post-publish verification does not claim registry
smoke, eval/model/OpenCode evidence, PH superiority, or generated app quality
certification.

## Implementation-To-Finish Product Usability Smoke

External Smoke completed one alpha7 Windows SSH registry implementation-to-finish
proxy trial.

Invocation lesson:

- failing form: positional prompt plus unqualified model id
  `gpt-5.4-mini-fast` produced `EUNKNOWN: unknown error, read`;
- passing form: stdin prompt plus provider-qualified model id
  `openai/gpt-5.4-mini-fast` returned OK;
- fallback `openai/gpt-5.4-mini` also returned OK;
- treat this as a local invocation/wrapper route issue, not a Persona Harness
  product issue or proven provider outage.

Observed facts:

- package source: registry `alpha`, installed
  `persona-harness@0.3.9-alpha.7`;
- Windows SSH D-drive workspace:
  `D:\persona-harness-smoke-alpha7-finish-20260629-131050-retryutf`;
- archive rollup:
  `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/alpha7-opencode-debug-20260629-131050/RESULT.md`;
- Korean idea-first input was preserved;
- no-token rail completed draft, approve, split, next, continue, and check;
- closure planner surfaced stable JSON and first actionable blockers;
- OpenCode used the PH rail, implemented a Spring Boot/Gradle backend, filled
  implementation/review reports, ran report-filled commands, archived `req-1`,
  and handled the Java role read coverage blocker;
- final `workflow finish implement`: PASS.

Verification observations:

- `gradlew.bat test`, `bootJar`, and `build`: `BUILD SUCCESSFUL` after generated
  test import repairs;
- runtime smoke observed create/update/delete success and 404/400 responses;
- final `workflow check` WARN had no pending tickets;
- remaining WARNs were report-only/non-blocking.

Boundary:

- this is one product usability trial only;
- it is not eval, A/B, PH superiority proof, generated app quality
  certification, or a general reliability claim;
- backend-shape verification WARN remains a possible future evidence
  readability/report-only issue, not a finish blocker.