메뉴
Orchestrates the full development workflow: research → plan → implement, with human review checkpoints between each phase. Use this skill when starting to implement a new feature end-to-end.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | development-runner |
| description | Orchestrates the full development workflow: research → plan → implement, with human review checkpoints between each phase. Use this skill when starting to implement a new feature end-to-end. |
| context | main |
| model | opus |
| effort | low |
You are orchestrating a four-phase development workflow. Execute each phase in order, pausing for human approval before advancing to the next.
Use the Skill tool to invoke the researcher skill. Pass the feature/topic from the user's original request as context.
After researcher completes, follow the Annotation Review procedure below, then move to Phase 2.
Use this procedure after Phase 1 and Phase 2 complete. Run the following loop until the user approves.
AskUserQuestion to tell the user: "Please review the output. If you have any feedback, add // ANNOT: <your comment> annotations to the document, then let me know. If everything looks good, reply 'approve'."annotation-fixer skill to apply the annotations, then return to step 1.Use the Skill tool to invoke the my-planner skill.
After my-planner completes, follow the Annotation Review procedure above, then move to Phase 3.
Use the Agent tool with subagent_type: task-decomposer to launch the task-decomposer agent. Pass the path to the implementation plan (.claude/user/plan/) as context.
After the decomposer subagent completes, use the Agent tool with subagent_type: task-runner to launch the task-runner agent. Pass the path to the task graph (.claude/user/plan/task-graph.md) as context.
After all task agents complete, note the worktree branch returned in the task-runner result. Run git checkout <branch> via the Bash tool to switch the current working directory to that branch before proceeding to Phase 4.
Run the following loop until the code-reviewer returns APPROVE:
Agent tool with subagent_type: code-reviewer to review the current changes.Once approved, proceed to Phase 5.
Use the Agent tool to spawn a subagent with the following prompt:
Compare the actual implementation on the current branch against the documents produced in earlier phases:
- Research document:
.claude/user/research/- Implementation plan:
.claude/user/plan/Identify any discrepancies — features added, removed, or changed during implementation that are not reflected in the documents. For each discrepancy found, update the relevant document to match the actual implementation.
Report a summary of what was updated.
After the subagent completes, inform the user the workflow is complete.
Use when the user asks for a rich explanation of a code change, diff, branch, or PR. Produces HTML output.
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
Control herdr from inside it. Manage workspaces and tabs, split panes, spawn agents, read output, and wait for state changes — all via CLI commands that talk to the running herdr instance over a local unix socket. Use when running inside herdr (HERDR_ENV=1).
Monitor a PR's CI, re-arm on each iteration, ignore known-flaky errors, stop when green
Reads annotations left by the user and updates the entire document based on their content.
Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch.