一键导入
impl
Implement a change with a pipeline scaled to its complexity. Single entry point for all code changes — auto-detects effort or accepts small|mid|tuff override.
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
Implement a change with a pipeline scaled to its complexity. Single entry point for all code changes — auto-detects effort or accepts small|mid|tuff override.
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
| name | impl |
| description | Implement a change with a pipeline scaled to its complexity. Single entry point for all code changes — auto-detects effort or accepts small|mid|tuff override. |
Single entry point for all code changes. Select effort explicitly or let Claude auto-detect from the request.
$0 -- (optional) Effort tier override: small, mid, or tuff. Omit to auto-detect.| Tier | Scope | Pipeline |
|---|---|---|
small | 1–2 files, modifying existing logic, no new abstractions | Read → 1 question max → implement → just check |
mid | Existing pattern, chain of /add-* skills, 3–8 files | Quick research → chain /add-* skills in order → just check |
tuff | New domain, new abstractions, or no prior pattern to follow | Full pipeline: questions → research → plan → plan-review → user-approval → implement → verifier |
Choose tuff if any of:
/add-* skill covers the taskChoose mid if any of:
/add-* skills directly match the taskChoose small if all of:
When ambiguous between mid and tuff, choose mid. When ambiguous between small and mid, choose mid.
just check. Fix any issues before reporting done.No planning. No agents. No approval step.
/add-* skills apply (e.g., /add-handler + /add-endpoint + /add-test)./research <domain> to confirm existing patterns — never invent structure when an example already exists./add-* skills in dependency order: entity before repo, repo before handler, handler before controller, then tests.just check. Fix all issues.No plan file. No approval step.
Follow the full workflow:
/research <topic> — read all relevant existing code thoroughly/plan <feature> — produce the structured plan with every file, field, method signatureExitPlanMode to present the plan and wait for approval. Outside plan mode, skip — proceed directly to implementation.just check after each major step; do not deviate from the plan without askingRun the plan-reviewer agent only when the user explicitly asks for a plan review.
Add a complete REST endpoint from command to controller route. Use when building new API endpoints.
Summarize the API interface changes made during THIS session for the frontend, then optionally spawn an agent to implement them in the frontend repo. Use after changing endpoints, request/response DTOs, or auth.
One-shot guided session — prunes unused code, scaffolds new entities/endpoints, updates project identity config, and replaces the init migration for a specific project.
Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md. Trigger: /caveman-compress FILEPATH or "compress memory file"
Run a Linear task end-to-end — creates an isolated tmux session + worktree, starts a Claude agent to implement it, opens a PR, and watches for review feedback automatically.
Add a new domain event with its publisher call and background event handler. Use for async side effects.