# v0.4.0-rc.1 Release Notes ## Version `v0.4.0-rc.1` ## Dist Tag `next` `0.4.0-rc.1` is a workflow closure rail product milestone candidate. Publish it with npm dist-tag `next`. Do not move `latest`; `latest` is currently observed at `0.3.9-alpha.8`. ## Summary This release candidate refreshes the previously parked RC draft from current alpha8 state. It packages the closure rail convergence that routes active workflow guidance and finish blockers through closure state/payload. Included closure rail state: - `workflow continue` uses the closure rail as its single source, without old follow-up dual-rendering; - `workflow finish implement` blocker/reason rendering uses closure payload/state via `workflowClosureFinishReasons`; - `workflow closure status --json` remains a state snapshot; - `workflow closure next --json` includes `nextStep` and first actionable blocker details; - old string layers `workflow-post-build-closure.ts`, `workflow-continue-followups.ts`, and `workflow-finish-reasons.ts` are retired. Not included: - no `closure run` executor; - no report auto-fill; - no ticket auto-archive; - no workflow auto-finish; - no finish gate weakening. ## Product Usability Signal Alpha8 registry smoke provides the release-candidate evidence base: - no-token closure rail smoke passed from registry `persona-harness@alpha` / `0.3.9-alpha.8`; - Korean idea-first input was preserved; - `workflow continue`, `workflow closure status --json`, `workflow closure next --json`, and `workflow finish implement` shared the closure blocker order; - implementation-to-finish product usability smoke passed once from registry `persona-harness@alpha`; - final `workflow finish implement` passed under closure-state routing; - terminal closure state had `currentTicket: null`, no pending tickets, filled reports, verification passed, and empty blockers. Boundary: - one product usability success-path smoke only; - not eval/A-B evidence; - not PH superiority proof; - not generated app quality certification; - not a closure guarantee; - not a general reliability guarantee. ## Eval Status Eval remains stopped. `docs/current/injection-value-status.json` remains `injection-effect-not-proven`. This RC must not reinterpret old eval results, claim PH superiority, or use generated app output as product-quality certification. ## Verification 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 / 464 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@0.4.0-rc.1`. - filename: `persona-harness-0.4.0-rc.1.tgz`; - package size: `395.7 kB`; - unpacked size: `1.6 MB`; - shasum: `a78f331ed4f1bb85506cad44e9041185c4f370d0`; - total files: `379`. - `git diff --check`: passed. ## Publish Notes - Preferred publish command: `npm publish --tag next`. - Do not move `latest`. - If npm requires OTP/2FA, stop after the first failure and hand off the exact command for the user to run. - Create and push `v0.4.0-rc.1` only after registry verification confirms `persona-harness@0.4.0-rc.1` exists with the release prep commit as gitHead. ## Post-Publish Verification User-completed publish was verified on 2026-06-29. Registry facts: - package: `persona-harness@0.4.0-rc.1`; - gitHead: `981b7b75d035ad16ee5e2b7f8bec5482d19f2873`; - dist.shasum: `c987701b4ed55324e1a555f4e31545e4d0cb3757`; - dist-tags: `next=0.4.0-rc.1`, `alpha=0.3.9-alpha.8`, `latest=0.3.9-alpha.8`. Git facts: - local and origin tag `v0.4.0-rc.1` point at `981b7b75d035ad16ee5e2b7f8bec5482d19f2873`; - local `main` and `origin/main` include the release prep commit. Boundary: - no republish was performed; - `latest` was not moved by this post-publish verification; - no eval/model/OpenCode run was performed; - no PH superiority proof, generated app quality certification, closure guarantee, or general reliability guarantee is claimed. ## `@next` Real-User Smoke External Smoke verified `persona-harness@0.4.0-rc.1` from registry `next`. - Stage 1 no-token closure rail smoke: PASS in a stable local workspace. - Stage 2 local implementation-to-finish proxy trial: PASS. - OpenCode invocation used stdin plus `openai/gpt-5.4-mini-fast`. - Generated app shape: Java/Spring/Gradle Todo API. - `gradlew test` / `gradlew build`: PASS. - Implementation and review reports were filled. - `req-1` was archived. - `workflow closure next` reached terminal complete state. - final `workflow finish implement`: exit 0. - `closure run` remains deferred because this local trial finished with the planner / continue / finish surfaces only. - Windows SSH remote validation was BLOCKED by SCP/SSH hang before product behavior. Do not count it as Windows product PASS or FAIL. Windows local operator validation: - source note: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/rc1-windows-validation-path-20260629-173647/user-local-operator-pass-20260629/RESULT.md`; - source transcript: `/Users/yongtae/.codex/attachments/19c17ce4-6035-418d-8fe3-6abadff31797/pasted-text.txt`; - workspace: `D:\persona-harness-smoke-rc1-next-local-20260629-174728`; - install source/version: registry `persona-harness@next`, installed `0.4.0-rc.1`; - setup PASS: init/default backend, doctor, bootstrap; Runtime readiness PASS; - Korean preservation PASS: `TODO 웹 서비스 만들래` was preserved in requirements, backlog, task card, continue output, and closure JSON; - closure planner / finish strictness PASS: - first blocker: `verification-unknown`; - expected no-model finish block: `FINISH_EXIT=1`; - blockers: `verification-unknown`, `implementation-report-missing`, `review-report-missing`, `evidence-missing`, `pending-ticket`; - no auto mutation: implementation/review reports stayed `Status: template`, the ticket stayed pending, and there was no auto-fill, auto-archive, or gate weakening. Proxy feedback from existing trial artifacts: - archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/rc1-next-real-user-20260629-164520/proxy-feedback-20260629`; - source trial archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/rc1-next-real-user-20260629-164520`; - workflow rail helpful signal: triggered; - `closure status` -> `closure next` -> `workflow finish` helped return to terminal/complete state without guessing; - decisive state cues: `currentTicket: null`, `pendingTickets: []`, `finish: passed`, `nextStep: terminal`; - commands-too-many: not triggered; - closure confusing: tolerable; - closure run requested: not triggered. Soak plan: - keep `0.4.0-rc.1` soaking under `@next`; - collect real external user feedback before stable `0.4.0`; - unblock/stabilize the Windows validation path, or use an alternate reliable Windows verification route; - Windows no-model closure surface route is now satisfied through direct local PowerShell operator validation; - SSH/SCP remote route and SSH command transport remain unsuitable for stable evidence because of instability/mojibake risk; - Windows current-head `002359c` focused SSH continuation was exercised but is PARTIAL/BLOCKED, not an implementation-to-finish PASS: - archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/current-head-002359c-ssh-focused-continuation-20260629-200215`; - the existing Windows D workspace was reachable over SSH and the prior Korean/stdin issue was no longer the blocker; - timeout alone was not the root cause; wrapper props were absent, but custom `gradlew.bat` could run; - initial Gradle failure was missing `junit-platform-launcher`; repair added `testRuntimeOnly 'org.junit.platform:junit-platform-launcher'`; - after repair, bearshell `gradlew.bat test` and build passed; - backend-shape generated with major shape checks mostly PASS, while DTO boundary and Verification report WARN remain; - source scan did not observe fake `gradle-shim`, Java `HttpServer`, CommonJS, or Express bypasses; - OpenCode closure continuation hung during the build bearshell step, reports stayed template, `req-1` stayed pending, and final `workflow finish implement` exited 1; - product learning: Windows can reach repair build/test PASS, but closure/report/archive/finish follow-through remains a blocker; future Windows validation should include a less script-heavy TUI/operator flow because the current automation/PowerShell/SSH route may overfit validation mechanics; - that less-script-heavy Windows operator route is itself BLOCKED before model until the guide/current CLI surface is corrected and reverified: - archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/less-script-heavy-windows-operator-exec-20260629-203000`; - a fresh D workspace installed current tarball `0.4.0-rc.1`; init, doctor, and bootstrap passed; - Korean preservation failed even with `Get-Content -LiteralPath idea-utf8.txt -Raw -Encoding UTF8 | npx ph workflow draft --stdin`; requirements `Original idea` became `??? ? ? API ???`; - the guide expected `workflow approve requirements`, `workflow split`, and `workflow next`, but the installed CLI reported unknown command for those in the run; - no OpenCode/model implementation executed; - final no-model finish exited 1 with expected blockers; - current HEAD `e688d39 fix(cli): guard lossy Windows stdin and pack stale dist` addresses those two current-head/fresh-tarball blockers for a future package: - pure question-mark replacement input such as `??? ? ? API ???` is rejected by `workflow draft --stdin` / `workflow capture --stdin` before requirements are written; - the existing `媛...???` mojibake guard remains and normal Korean UTF-8 stdin remains preserved; - `prepack: npm run build` prevents local smoke tarballs from carrying stale `dist`, which caused the guide/current CLI mismatch where `workflow approve requirements`, `workflow split`, and `workflow next` appeared missing; - this does not reconstruct already-lost `?` text, does not repair existing broken artifacts, and is not proof that published `0.4.0-rc.1 @next` already contains the fix; - current-head `e688d39` Windows operator retry confirmed preflight progress but remained BLOCKED before app output: - archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/current-head-e688d39-windows-operator-retry-20260629-204500`; - fresh current HEAD tarball with `e688d39` installed on Windows D-drive; - packaging/stale `dist` issue was resolved and current command surface was available; - lossy `??? ? ? API ???` stdin was rejected without writing requirements; - Korean `간단한 할 일 API 만들래` was preserved when PowerShell encoding variables were set to UTF-8; - no-token workflow draft/approve/split/next/continue/check/closure status/next passed; - implementation-to-finish blocked before app output: OpenCode started and read workflow/profile/planner context, README was absent, and a malformed duplicated `.persona/policies` path hit `external_directory` auto-reject; - no `src` or Gradle files were generated, reports stayed template, ticket stayed pending, and final `workflow finish implement` exited 1; - blocker moved from Windows input/packaging/command surface to OpenCode implementation entry/context path; stable `0.4.0` remains deferred; - current HEAD `a307ac0 fix(cli): guide README-absent workflow entry` addresses the CLI-owned part of that entry/context ambiguity: - CLI investigation did not find the malformed duplicated `.persona/policies` absolute path in CLI/operator prompt output, so it is likely model/operator generated; - README-absent workspaces still had rail guidance that could over-emphasize README or `.persona/policies` directory reading; - `workflow implement`, `workflow continue`, and `plan --implement` now explicitly handle README absence and guide agents to repo-relative source-of-truth files: `.persona/project-profile.jsonc`, `.persona/policies/overlay.jsonc`, `.persona/workflow/plan.md`, and the current ticket / requirements source; - README-present behavior keeps README chunk guidance; - this reduces entry/context ambiguity but does not guarantee model follow-through, weaken finish gates, auto-fill reports, or auto-archive tickets; - current-head `a307ac0` Windows operator retry confirmed preflight progress but remained implementation-to-finish PARTIAL/NOT PASS: - result: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/current-head-a307ac0-windows-operator-retry-20260629-212905/RESULT.md`; - stale `dist` was fixed, approve/split/next command surface was available, lossy `???` stdin was rejected, Korean was preserved, no-token workflow/closure preflight passed, README-absent guidance was observed, and the malformed duplicated `.persona/policies` path did not recur; - OpenCode read `.persona/policies/overlay.jsonc`, but no app output was generated and `src` / Gradle files were absent; - the model still filled reports/report-filled and archived `step-1`; - final `workflow finish implement` exited 1 on `STACK_MISMATCH` plus pending `step-2`, `step-3`, and `step-4`; - interpretation: previous Windows input/packaging/README-absent path blockers improved or resolved, while the new blocker is report/archive integrity before final finish. The finish gate correctly blocked, but report-filled/archive steps may be too permissive before real implementation/evidence. Stable `0.4.0` remains deferred; - current HEAD `8660ef3 fix(cli): guard ticket archive on closure blockers` addresses the archive side of that integrity blocker: - `plan --report-filled` remains a report marker, not a product-quality gate; - `workflow archive ` now reads closure state and refuses to move work to history when non-ticket closure blockers remain; - archive-blocking non-ticket blockers include `verification-unknown`, `evidence-missing`, `report-coverage-missing`, and stack / verification / report blockers; - pending-ticket and history-repair blockers are not archive blockers; - if reports are marked filled but no app/evidence/verification exists, archive exits 1 and leaves backlog/work ticket pending; - no report auto-fill, no auto-archive, and no finish gate weakening is introduced; - current-head `8660ef3` Windows operator retry then classified as INTEGRITY-PASS/PARTIAL: - archive: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/current-head-8660ef3-windows-operator-retry-20260629-221043/RESULT.md`; - fresh tarball preflight passed: prepack/build, approve/split/next/archive command surface, lossy `???` rejection, Korean UTF-8 preservation, README-absent guidance, and no-token workflow/closure preflight; - no-app/no-evidence `workflow archive step-1` exited 1 with `verification-unknown`, `implementation-report-missing`, and `review-report-missing`, and `workflow next` kept `step-1` current; - after OpenCode generated a minimal Java/Spring/Gradle skeleton and evidence made verification passed, `step-1` moved to history; - final finish still exited 1 because `step-2`, `step-3`, and `step-4` remained pending, with backend-shape WARNs remaining; - interpretation: integrity guard verified; single ticket archived after evidence; finish blocked correctly due pending tickets, so stable `0.4.0` remains deferred; - current HEAD `0784135 fix(cli): block controller repository closure` adds the first scoped hard convention blocker: - existing observer signal `controller.repository-dependency` is promoted to hard workflow closure blocker `architecture-controller-repository-direct-dependency`; - applies only to ready Java/Spring service-layer profile styles and only when typed evidence names the Controller, Repository, source file, and direct dependency; - `workflow check`, `workflow closure next --json`, `workflow continue`, `workflow finish implement`, and `workflow archive ` surface/block this violation consistently; - action wording tells the agent to route the Controller through a Service layer instead; - compliant Controller -> Service -> Repository paths have no architecture blocker; - other backend-shape WARNs remain report-only; this is not broad architecture correctness enforcement and does not add `closure run`, custom conventions, report auto-fill, auto-archive, or finish gate weakening; - current HEAD `7fda771f74008f42082c3a85377262c8fc7ccf5f` has GUARD Phase 0-3 QA PASS and package-surface smoke PASS on a fresh current tarball, not the published `0.4.0-rc.1 @next` package: - included commits: `c2819ac fix(cli): add opt-in direct verification gate`, `ffeafa9 fix(cli): support convention blocker levels`, `2f561a0 fix(runtime): warn on write-time convention violations`, and `7fda771 refactor(cli): add convention registry`; - QA verification: focused 134 tests PASS, full `npm test` 70 files / 486 tests PASS, typecheck PASS, build PASS, product smoke PASS, and built CLI smoke PASS; - package-surface smoke tarball: `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/current-head-7fda771-guard-package-surface-smoke-20260630-001558/pack/persona-harness-0.4.0-rc.1.tgz`; shasum `b703953aab409f1cd7ac578c5af76b3d3e42cf90`, sha256 `2155ac28c48367c85d2a4163ba56ecea5dd1842b1d0e0935313c665ab9d55b7c`; - Phase 0 PASS: opt-in `enforce.executeVerification: true` direct verification gate for the supported Java/Spring/Gradle slice. Fake agent-written passed evidence no longer passes; PH-run direct `gradlew test` / JUnit success passes; JUnit failure blocks. Disabled mode keeps the existing structured bearshell evidence flow; - Phase 1 PASS: convention levels `report|warn|block`; the scoped `controller.repository-dependency` block level creates `architecture-controller-repository-direct-dependency`; warn/report do not hard-block and compliant Controller -> Service -> Repository has no blocker; - Phase 2 PASS/PARTIAL by design: write-time convention guard is warning-only because hook deny/rewrite is not supported by the current plugin hook result type. Do not call this hard enforcement; - Phase 3 PASS: convention registry aligns id/default level/blocker/fix path across observe, check, closure, continue, and write-warning surfaces. BYO `.persona/conventions/*.yml` ast-grep authoring remains future work; - boundaries: scoped product enforcement behavior only; no eval/A-B proof, PH superiority, generated app quality certification, broad architecture correctness, general reliability, or closure guarantee; - post-rc1 current HEAD includes `002359c fix(cli): reject lossy Windows stdin mojibake` after the rc1 Windows implementation trial was blocked before the model by an ASCII-safe script using `Get-Content -Raw | npx ph workflow draft --stdin`; PowerShell 5.x can read UTF-8 no-BOM files as ANSI/CP949 first and produce already-lossy text such as `媛꾨떒??????API 留뚮뱾?`; - for future Windows file-pipe trials, use `Get-Content -LiteralPath -Raw -Encoding UTF8 | npx ph workflow draft --stdin` and the same `-Encoding UTF8` form for `workflow capture --stdin`; the new guard rejects lossy mojibake nonzero instead of writing corrupted requirements, but it cannot repair existing broken artifacts and is not in the published rc1 registry package until a future package includes it; - watch report consistency/noise during soak: `verification passed` plus report-only WARN/backend-shape split can confuse users; - keep `closure run` deferred unless repeated same-step friction appears; - do not move to stable `0.4.0` until those preconditions are addressed. Boundary: - product usability smoke only; - no eval/A-B evidence; - no PH superiority proof; - no generated app quality certification; - no closure guarantee; - no general reliability guarantee; - no `latest` move.