一键导入
spec-writing
Guide for creating high-quality specifications for sandbox. Use when starting a new SPEC or reviewing specs against architectural standards.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Guide for creating high-quality specifications for sandbox. Use when starting a new SPEC or reviewing specs against architectural standards.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Advanced `auto-continue-pr` workflow for PRs started by `auto-create-pr-loop`. Claims the PR, re-enters an isolated worktree, resumes from the first non-done row in `.ai/runs/<date>-<slug>/PLAN.md`, executes lean per-step commits, batches verification into `checkpoint-<N>-checks.md` every 5 resumed steps (with focused integration tests + screenshots when UI was touched), runs the full validation gate plus full/standalone integration suites and ds-guardian at spec completion, and preserves the run-folder and label contract. Use the original `auto-continue-pr` for simple `auto-create-pr` runs.
Resume an in-progress pull request that was started by the auto-create-pr skill. Given a PR number, claim the PR under the in-progress lock protocol, check its branch out into an isolated git worktree, locate the execution plan linked from the PR body, read its Progress checklist, and continue execution from the first unchecked step with incremental commits and progress updates until the PR is complete. Runs the same validation gate (typecheck, unit tests, i18n, build) and label discipline as auto-create-pr. Usage - /auto-continue-pr <PR-number>
Advanced `auto-create-pr` workflow for long, multi-step spec implementations that need resumability and strict step tracking. Creates a run folder under `.ai/runs/<date>-<slug>/` with `PLAN.md`, `HANDOFF.md`, and `NOTIFY.md`, executes one lean commit per task-table step, batches verification into `checkpoint-<N>-checks.md` every 5 steps (with focused integration tests + screenshots when UI was touched), runs the full validation gate plus full/standalone integration suites and ds-guardian at spec completion, and opens a PR with the correct labels. Use the original `auto-create-pr` for small fixes.
Execute an arbitrary autonomous agent task end-to-end and deliver it as a GitHub pull request against develop. Start by drafting an execution plan in .ai/runs/ that includes a Progress checklist, commit it on a fresh task branch in an isolated worktree, implement the work phase-by-phase with incremental commits, update the Progress checklist after every phase, optionally honor one or more external reference skills passed by URL, run the full validation gate (typecheck, unit tests, i18n, build) for any code changes, and open a PR with the correct pipeline labels. Resumable via the auto-continue-pr skill.
Fix a GitHub issue by number from the current repository. First check whether the issue is already solved or already has an open solution, then use an isolated git worktree to implement the minimal fix, add unit tests, run code review and backward-compatibility checks, run validation including i18n, typecheck, unit tests, and other required checks, then push a branch and open a pull request with a full description linked to the original issue.
Review or re-review a GitHub pull request by number in an isolated git worktree. Fetch the specific PR from GitHub, run the full code-review skill, submit approve or request-changes, manage labels, and if blockers remain offer an optional autofix and fix-forward flow that iterates through conflict resolution, code fixes, unit tests, typecheck, and re-review until the PR is merge-ready or a real blocker remains. Usage - /auto-review-pr <PR-number>
| name | spec-writing |
| description | Guide for creating high-quality specifications for sandbox. Use when starting a new SPEC or reviewing specs against architectural standards. |
Design and review specifications (SPECs) against Open Mercato architecture and quality rules.
AGENTS.md for module conventions and .ai/specs/ for existing specs.{date}-{title}.md in .ai/specs/.Q1, Q2, …) after the TLDR.Use the Specification Template. Adapt if needed, but ensure core concerns are addressed.
Required sections: TLDR, Problem Statement, Proposed Solution, Data Models, API Contracts, Risks, Changelog.
# Architectural Review: {SPEC-0XX: Title}
## Summary
{1-3 sentences: what the spec proposes and overall health}
## Findings
### Critical
{Cross-module ORM, tenant isolation leaks, missing auth guards}
### High
{Missing undo logic, incorrect module placement, missing phase strategy}
### Medium
{Missing failure scenarios, inconsistent terminology}
### Low
{Style suggestions, nits}
makeCrudRoute, <CrudForm>, <DataTable>, apiCall / useGuardedMutation, DI-resolved cache, createModuleEvents) or invent its own substitute? See AGENTS.md → Mandatory Module Mechanisms for the full canon and links. No raw fetch, no raw <form>, no new Redis(...), no manual cross-module ORM joins.encryption.ts defaultEncryptionMaps entry and route reads through findWithDecryption? See AGENTS.md → CRITICAL Rule #11 (Encryption maps) + the "Encryption maps for sensitive data" row of the Mandatory Module Mechanisms table and .ai/skills/data-model-design/SKILL.md § Sensitive Data and Encryption Maps. No hand-rolled AES, no crypto.subtle, no "TODO encrypt later".text-red-* / bg-green-*), Tailwind text scale (no text-[11px] / text-[13px]), shared primitives (StatusBadge, Alert, FormField, SectionHeader, CollapsibleSection, LoadingMessage / Spinner / DataLoader, EmptyState), lucide-react icons in page body (never inline <svg>), dialog Cmd/Ctrl+Enter submit + Escape cancel, aria-label on every icon-only button? See AGENTS.md → CRITICAL Rule #10 (Strict Design System alignment for every UI change) and .ai/skills/backend-ui-design/SKILL.md. Specs that touch existing pages MUST honour the Boy Scout rule.organization_id is mandatory for all tenant-scoped entities.<module>/encryption.ts, read via findWithDecryption) — see AGENTS.md → Data Encryption.makeCrudRoute), backend forms (CrudForm), tables (DataTable), HTTP (apiCall — never raw fetch), non-CrudForm writes (useGuardedMutation), cache (DI-resolved @open-mercato/cache), events (createModuleEvents) — see AGENTS.md → Mandatory Module Mechanisms.<svg> in page-body UI. See AGENTS.md → Design System..ai/skills/data-model-design/SKILL.md → Sensitive Data and Encryption Maps.ai/skills/module-scaffold/SKILL.md → Encryption maps.ai/skills/backend-ui-design/SKILL.md — DS-compliant pages