القائمة
Final Smelter human approval gate for evidence review, completion decisions, commit approval, or rework routing.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when: User wants to run a SUBSTANTIAL multi-step coding task as a workflow (fix/implement a whole issue or feature, comprehensive review), CREATE workflows or commands, set up workflow automation, or manage workflow configuration. Triggers (run): "use workflow to", "run workflow", "with workflow", "workflow to", "workflow run", "use a workflow to", "run a workflow", and generic task-completion intents that imply a full coding deliverable: "실행해줘", "돌려줘", "구현해줘", "고쳐줘", "이슈 고쳐줘", "PR 만들어줘", "implement this", "fix this issue", "build this feature", "ship this". Triggers (create): "create a workflow", "write a workflow", "make a command", "author a workflow", "new workflow", "new command", "workflow yaml". Triggers (setup): "set up workflows", "install workflow automation", "how to use workflows", "configure workflows", "workflow setup", "get started with workflows". Triggers (config): "change my workflow config", "modify workflow config", "workflow
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
Smelter deep interview for planning-first requirement discovery before /brainstorm, /implement, or /fix execution
Create a release from dev branch. Generates changelog entries from commits, bumps version, and creates a PR to main. TRIGGERS - Use this skill when user says: - "/release" - create a patch release (default) - "/release minor" - create a minor release - "/release major" - create a major release - "make a release", "cut a release", "ship it", "release to main"
Replicate and validate a GitHub issue by spinning up Smelter, analyzing the issue, and systematically testing all described symptoms using browser automation. Use when: User wants to reproduce a bug, validate a GitHub issue, confirm a reported problem, or investigate whether an issue is real before working on a fix. Triggers: "replicate issue", "reproduce issue", "validate issue", "confirm bug", "test issue", "can you reproduce", "try to replicate", "verify the bug". Capability: Checks out main, pulls latest, starts Smelter, reads the GitHub issue, then uses agent-browser to systematically test every symptom and produce a findings report. NOT for: Fixing issues (use /smelter or /exp-piv-loop:fix-issue), general UI testing (use /validate-ui).
Autonomous rule adherence checker. Scans the codebase for rule violations, fixes the highest-impact ones in an isolated worktree, runs full validation, creates a PR. Uses memory to track progress across runs.
| name | workflow-human-check |
| description | Final Smelter human approval gate for evidence review, completion decisions, commit approval, or rework routing. |
| version | 0.55 |
| type | workflow |
| consumes | all artifacts + team_review.md |
| produces | user decision (rework | complete-commit-current | complete-new-branch-pr | hold | upgrade) |
| default_pattern | User |
| supports_patterns | ["User"] |
| halts_session | true |
| gate | {"postcondition":[{"user_decision":"rework | complete-commit-current | complete-new-branch-pr | hold | upgrade"}]} |
Final user review. This is the only allowed halting point. No task is complete until the user chooses one of the complete-* actions here.
Announce: I'm using workflow-human-check to render the summary report and collect the user's complete / rework / hold / upgrade decision.
Use native AskUserQuestion for the decision. Plain-text menus such as Decide: complete / rework / hold / upgrade? are invalid and will be rejected by the Stop hook.
If the runtime has not loaded the deferred tool schema yet, first load it with ToolSearch("select:AskUserQuestion"), then invoke AskUserQuestion. Do not fall back to prose unless the user explicitly says native tools are unavailable.
<N files changed, +X/-Y lines> plus top paths.Read before asking: png/jpg/gif/webp, video representative frame/path, pdf.Write a concise Markdown report and results.md under the canonical task directory. Use tables for scannability.
Summary table:
| Field | Value |
|---|---|
| Feature | <slug> |
| Task | <title> - user-facing before/after or concrete failure fixed |
| Mode | <mode> |
| Stages | <completed workflow skills> |
Artifacts table:
| Kind | Path / Summary |
|---|---|
| Visual | <path> / (none) / MISSING - required |
| Log | <path> / (none) / MISSING - required |
| Diff | <N files changed, +X/-Y lines> · top 3 paths · +M more |
Verification table:
| Check | Result |
|---|---|
| TDD/typecheck | <pass/fail + command> · `scope: scoped |
| E2E | <surface + pass/fail + artifacts> · `scope: scoped |
Risks: bullets only, severity [LOW|MEDIUM|HIGH|CRITICAL], omit if empty.
Before asking for completion, read verify_report.md, agent-review.md, e2e-review.md, team_review.md, and results.md when present. Treat these as unresolved findings: any fail verdict, missing required artifact/evidence, unchecked acceptance item, MEDIUM|HIGH|CRITICAL risk, or a user-visible blocker such as a deploy gate.
If unresolved findings exist, summarize them first with file/artifact anchors, then ask one problem-focused AskUserQuestion. Do not offer complete-commit-current or complete-new-branch-pr in this question. Use exactly these options and no free-form catch-all:
rework - create active_feedback, route to the producer skill named by the finding.hold - record blocked/paused status with the unresolved findings.upgrade - move to /fix, /implement, or /brainstorm when the finding needs a stronger mode.This applies to every mode, including /verify after a previous human-check. A fresh verification that finds missed work must ask about the missed work, not present commit choices.
Only when there are no unresolved findings, ask one native AskUserQuestion with exactly these options and no free-form catch-all:
rework - create active_feedback, route to target skill.complete-commit-current - finalize, then git commit on the current branch.complete-new-branch-pr - finalize, create a new branch, commit, push with upstream, then run gh pr create.hold - record blocked/paused status.upgrade - move to a stronger mode and re-enter the proper skill.Do not paraphrase option labels. Do not ask a free-form “continue?” question. Do not ask a second git-options question. Do not render the options as Markdown/plain text; the decision must be collected through native AskUserQuestion.
On either complete-* action, write non-empty results.md. finalize-human-check.mjs observes that write for fix, implement, and infra, then records the workflow-human-check pass event and completed_stages entry. Do not edit .state.json yourself.
For complete-commit-current, after finalize evidence exists:
git status --short.git commit with the repository's <type>: <description> style.git status --short to verify.For complete-new-branch-pr, after finalize evidence exists:
git status --short.git commit with the repository's <type>: <description> style.gh pr create with a concise title and summary.Never commit before the finalize pass shape exists. Never push, PR, merge, delete branches, or discard unless the user explicitly chooses complete-new-branch-pr or separately requests it.
| Decision | Next action |
|---|---|
complete-commit-current | write results.md, let finalize hook pass, then git commit locally |
complete-new-branch-pr | write results.md, let finalize hook pass, create/switch to a new task branch, push, then gh pr create |
rework | add unresolved feedback and route to target skill |
hold | stop with blocked/paused status |
upgrade | transition mode and continue |