菜单
Execute phase of the workflow chain. Uses Haiku (simple), Sonnet (standard), or Opus (complex) to implement code changes according to a plan. Use when the user has an approved plan ready to execute.
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
Clarify phase of the workflow chain. Uses Opus to analyze the user's request, resolve ambiguities, define scope, and produce a clear task brief before research begins.
Flow workflow chain that runs clarify → [research] → plan → execute → validate end-to-end without stopping. Pauses only on genuine ambiguity or circuit-breaker conditions.
Plan phase of the workflow chain. Uses Opus to design an implementation strategy. Use when the user wants to create a detailed plan before executing changes.
Research phase of the workflow chain. Uses Haiku (simple) or Sonnet (standard) to research the codebase and gather context for a task. Use when the user wants to investigate code before planning.
Validate phase of the workflow chain. Uses Sonnet to review and validate implementation changes. Use when the user wants to validate that code changes are correct and complete.
Compressed 3-phase fast-feedback workflow chain (clarify → execute → validate) with cap-3 parallel executors, no retry loop, fail-fast. For prototypes, spikes, and quick refactors.
| name | st-execute |
| description | Execute phase of the workflow chain. Uses Haiku (simple), Sonnet (standard), or Opus (complex) to implement code changes according to a plan. Use when the user has an approved plan ready to execute. |
Implement the code changes for the user's task.
steop state set-phase execute --mode execute
Launch the executor agent (steop:executor, full tool access). Override model based on task complexity:
model: "haiku"model: "sonnet"model: "opus"If the plan contains independent steps (changes to separate files or modules with no cross-dependencies), launch multiple Execute agents in parallel — one per independent group. Each agent should receive:
The execution agent(s) should follow the plan step by step and return as soon as the planned steps are complete. Prefer YAGNI — skip defensive code, edge cases, and polish unless they block the happy path or are named in the plan. Leave TODOs where assumptions are load-bearing. Do NOT refactor neighboring code. Only remove imports, variables, or functions that your changes made unused. Every changed line should trace to the plan. Report what was changed after completion.
After the agent(s) complete, summarize all changes made (files modified/created, key changes in each).