Gsx vrf autorun

gsx-vrf-autorun

Codex Adapter

In Codex, translate this Claude wrapper instead of rewriting it:

• $ARGUMENTS = text after $gsx-*.
• /gsd:<name> ... or Skill("gsd:<name>", args="...") = run .codex/skills/gsd-<name>/SKILL.md with args as {{GSD_ARGS}}.
• AskUserQuestion: only map to request_user_input if your Codex build truly renders an interactive picker; otherwise (the usual Codex case) do not force it. Before any choice among options/plans, first print each option's core content in your reply, then list the choices as a numbered Markdown list and stop and wait for the user's text reply. Never self-answer a prompt the user never saw.
• Claude tool names are intent labels; use equivalent Codex tools.
• Preserve wrapper boundaries; if GSD owns edits, do not edit source here.

Machine-pre-run the <how-to-verify> checklist of a GSD checkpoint:human-verify gate (gate="blocking"). Pre-step for gsx-vrf-approved (which releases the gate).

Boundary (hard rules):

• Verify only, never fix. Bug → report Gap section; don't change business code (fixes → /gsd:quick / /gsd:debug).
• Never release. Never reply <resume-signal>; on all-pass, prompt user to run gsx-vrf-approved.
• No writes / no edits to PLAN.md / STATE.md / SUMMARY.md. Report is conversation-only. No commits, no branch switches.
• Test data not force-cleaned (left for re-verification); report lists what was created.
• Reply in Chinese.

0 — Parse arguments, locate the target gate

Numbering: phase dirs .planning/phases/{NN}-{slug}/ (2-digit zero-pad); X.Y = Phase X / Plan Y; X.Y.Z = Phase X / Plan Y / Zth <task>.

$ARGUMENTS has a number — normalize (case / Phase·Plan·Task prefix / whitespace all optional):

User writes	Parsed as	Scope
4.4, Phase 4.4, Plan 4.4, 04-04	Phase 04 / Plan 04	All human-verify tasks in Plan
Plan 4, Plan 04	Plan 04 in current Phase (context/STATE.md)	Same
4.4.1, Task 4.4.1	Phase 04 / Plan 04 / 1st task	That task only (confirm human-verify)
Task 3	3rd task in active Plan	That task only

Locate via glob (zero-pad 2 digits; Plan may have letter suffix, e.g. 04-03a-PLAN.md):

ls .planning/phases/04-*/04-04*-PLAN.md

0 or multiple matches → list candidates, ask user, stop (don't guess).

$ARGUMENTS empty — infer:

1. Conversation: executor just returned a checkpoint:human-verify gate message? Its Plan is the target.
2. Else read .planning/STATE.md executing active phase; glob its *-PLAN.md; keep ones that have a checkpoint:human-verify task and no *-SUMMARY.md in the same dir.
3. Exactly 1 → use it. 0 or multiple → "Could not uniquely locate a pending checkpoint:human-verify gate — please specify a task / Plan number." and stop.

No .planning/ → not a GSD project; tell user and stop.

1 — Parse <how-to-verify> steps

Read all <task> of the Plan in document order; check each type.

Plan number → filter all type="checkpoint:human-verify" tasks, verify each. None → "Plan 0X-0Y has no human-verify task" and stop.

Task number → take the Zth task. human-verify → verify it. auto/other → "Task X.Y.Z is not a human-verify gate", point out the real human-verify task index, stop (don't silently switch).

For each gate task extract: <what-built> (feature scope accepted); <how-to-verify> (core: break into atomic steps, record each "expected observable result"); <resume-signal> (report footer hint only; never sent).

Also read same-dir {NN}-CONTEXT.md / {NN}-*-SUMMARY.md / *-UI-SPEC.md for: APIs/pages under test, credentials, initial data, env prerequisites. Root CLAUDE.md "Local Run" = authoritative cold-start guide.

2 — Classify each step

Category	Signals	Execution
Prep / cold-start	start backend/frontend, docker-compose, Flyway migration, service up	background tasks (step 3), confirm readiness
Web E2E	UI behavior: page/console/list/drawer/button/click/toast/form validation/badge	Playwright MCP
Pure API	HTTP contract: error codes, response fields, headers, /openapi/*, /admin/*, DB persistence	curl (+ mysql/redis-cli)
Skip	see below	skip

Skip when (can't machine-judge): subjective with no objective anchor (color/tint, "matches prototype", polish, layout, animation); prerequisite not self-constructible (real DID SDK, real upstream API, prod deploy/monitoring/backup, external systems); too vague (no success criterion); mixed step where automation covers only part → run coverable part, mark rest "partially skipped". When in doubt, prefer skip.

After classifying, show a brief table (gate task / step # / summary / category), then proceed.

3 — Prepare the environment

Probe first:

curl -s -o /dev/null -w "%{http_code}" http://localhost:8080/ ; curl -s -o /dev/null -w "%{http_code}" http://localhost:3000/

Steps require cold-start → stop first: docker-compose down (no -v — keep volumes), kill running dap-admin / frontend bg processes. Otherwise → start per below if down; reuse if online.

Follow CLAUDE.md "Local Run" via background tasks (run_in_background):

1. docker-compose up -d (MySQL 5.7 + Redis 7).
2. cd dap-server && mvn -pl dap-common install -DskipTests — install latest dap-common to ~/.m2 (foreground, wait).
3. mvn -pl dap-admin spring-boot:run — background (:8080). spring-boot:run cannot use -am.
4. cd dap-frontend && pnpm install && pnpm dev — background (:3000).
5. Poll: curl :8080 / :3000 until HTTP responds; check dap-admin logs for Flyway migration done, no real ERROR.

⚠️ Destructive ops need consent. Step needing "reset DB / delete docker/mysql/data or docker/redis/data" → AskUserQuestion first; declined → skip, note "DB reset required — not authorized". Local JDK must be 21; on Java version error switch JAVA_HOME per CLAUDE.md.

Services fail / migration errors → mark prep fail, record log excerpt, mark dependent steps blocked, stop trying.

4 — Execute steps

Run in <how-to-verify> order (data chains across steps — respect order).

Data seeding: self-seed, self-test. Login if needed (credentials from SUMMARY/CONTEXT; seed admin admin); "change password on first login" → walk it, note new password. Create prerequisite data (institutions/apps/scenes/templates) via API or UI.

4a — Prep / cold-start: start per step 3; confirm ready (migration version, ports, key pages load, no real ERROR). Record observed values.

4b — Web E2E (Playwright MCP): browser_navigate to target (:3000, /console, /login) → browser_snapshot (a11y tree) → browser_click/browser_type/browser_fill_form/browser_select_option → browser_wait_for on async (no blind sleep) → browser_snapshot to verify text/list/toasts, browser_console_messages for JS errors, browser_network_requests for API responses → browser_take_screenshot only to diagnose failures, browser_close at end.

Retry cap: ≤10 ops/step. Past 10 → blocked, record where it stuck, move on. Playwright MCP unavailable → mark all web steps blocked, note "Playwright MCP not ready — web steps need manual verification." Don't downgrade to reading code and guessing.

4c — Pure API (curl): assert HTTP status + R wrapper code/message/data. Headers: Open API → x-dap-appkey+x-dap-appsecret; Data API → x-dap-appkey+x-dap-token; Admin API → session. DB/Redis via mysql/redis-cli (t_audit_log, dap:token:*, dap:vc:revoked:*). Record request sent / received / matched.

Judgment: all expected items match → pass; any mismatch → fail (expected vs actual); env/selector stall + retries exhausted → blocked; can't objectively judge after running → skip (don't guess).

5 — Output summary report (conversation only)

Multiple gate tasks → one block each.

## Auto-Verification Report — Plan 0X-0Y

### Gate Task N · <task name>
Summary: total: M · pass: a · fail: b · skip: c · blocked: d

| # | Step | Type | Result | Notes |
|---|------|------|--------|-------|
| 1 | ... | Prep/Web/API | pass/fail/skip/blocked | actual result |

(repeat per human-verify task)

## Skipped Steps (manual follow-up)
- Task N Step K "<summary>" — reason

## Gaps (issues found)
- Task N Step M "<summary>" — fail: expected X, actual Y. Fix via /gsd:quick or /gsd:debug.
- (also blocked env + automation blind spots)

## Test Data Created
- institutions / apps / scenes / credentials / tokens created this run.

Report must call out: (1) skipped steps (which, why); (2) Gaps (fail/blocked + blind spots).

Closing prompt:

• All pass, no skips/Gaps → "Passed all steps — confirm, then run gsx-vrf-approved to release."
• Skips, no fail/blocked → "No issues, but N step(s) need manual follow-up — release after verifying those."
• Fail/blocked → "Gaps found — fix (via /gsd:quick or /gsd:debug) and re-verify before releasing."

Always remind: auto-verification ≠ human judgment — review skips first.