基于 SOC 职业分类
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/peel/fiddle --skill develop-loop命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Use after implementation to analyze drift, update documentation, evolve evaluation, and close a completed epic.
Use when starting or resuming a full development lifecycle for a feature or explicit epic.
Use when you have a spec or requirements for a multi-step task, before touching code.
| name | develop-loop |
| description | Use when a single task bean needs implementation and evaluation — called by fiddle:develop, not directly. |
Invoke as fiddle:develop-loop --bean <id> --epic <id>.
For every task, implement, gather evidence, evaluate each domain, merge scorecards, and iterate to a recorded terminal verdict. A shortened chain cannot turn an implementer claim into evidence.
Both --bean and --epic are required. Read evaluator configuration from orchestrate configuration: attended mode, dispatch budget, and per-domain template, providers, calibration, antipatterns, and thresholds.
Follow dispatch and evidence. It owns restart entry, evaluation-log initialization, active-marker lifecycle, domain resolution, implementer dispatch, evidence capture, runtime lifecycle, provider selection, evaluator dispatch, and scorecard validation.
Follow convergence and recovery. It owns scorecard merging, spec-defect exits, attended corrections, threshold and convergence verdicts, durable logging, re-dispatch behavior, and terminal states.