메뉴
Use when repeated project knowledge, user corrections, quality dissatisfaction, SOP patterns, project skill updates, AGENTS.md updates, templates, or eval cases should be captured and promoted. 当需要沉淀项目经验、踩坑、用户反馈、项目 skill、AGENTS.md、模板、测试夹具或自动迭代 skill 时使用。
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
Classify a new or materially changed software-development goal and submit its route decision to workflow-control-plane. Use at task intake when risk, strategy, SpecFlow/OpenSpec, context grounding, handoff, harness, or downstream skill selection must be decided; do not use for ordinary continuation when an active workflow manifest already selects the current stage skill. 当新开发目标进入、范围或风险实质变化、需要选择开发策略/Spec/上下文/交付流程时使用；已有 active workflow 的普通续作直接恢复当前 stage，不重复触发本 skill。
Coordinate role-based AI agents through agent rosters, work orders, context packets, artifact refs, and structured result contracts. Use when a software workflow needs multiple roles such as context researcher, spec writer, spec reviewer, technical designer, implementer, code reviewer, tester, verifier, or when the user wants an orchestrator to manage progress without manually copying context between sessions. 当需要多 Agent 协作、角色隔离、上下文投影、work order、artifact 驱动交接、进度监管或避免人工搬运上下文时使用。
Select and run changed-file-aware tests for software development loops, checkpoints, and completion gates. Use when a task should avoid repeatedly running the full unit suite, needs diff-based test impact selection, test cadence decisions, or must escalate from focused tests to broader verification. 当开发任务需要按 diff 跑增量测试、选择受影响单测、控制测试频率、避免每个 task 全量测试或判断何时升级为模块/全量验证时使用。
Owns workflow runtime state for adaptive AI coding tasks. Use when resolving route decisions, selecting versioned strategies, creating/updating workflow_manifest.json, validating artifact graphs, resuming interrupted workflows, applying specialist skill transition results, or enforcing strategy/stage/claim/state contracts. 当需要策略解析、workflow_manifest 单点写入、状态迁移、resume、artifact graph、stale 传播、claim ceiling 或工作流状态检查时使用。
Use when validating evidence manifests, deciding dev_done/integration_done/handoff_done claims, issuing verifier attestations, preventing completion overclaim, or requiring fresh consumer / real external verification. 当任务需要证据矩阵、交付验收、claim attestation、fresh consumer、real external、E2E/integration/system 证据或防止假完成时使用。
Initialize or repair a project AI coding harness: AGENTS.md, OpenSpec-first or Superpowers-fallback spec/design/plan routing, docs/evidence, Goal Loop Mode prompt, agent team roles, project-local skill, learning candidates, and delivery verification templates. Use when starting a new repo/product area, first MVP vertical slice, Large feature lacking current-truth docs, repeated user explanations, weak project agents, missing spec/technical-design/plan/evidence contracts, or when adaptive-dev-workflow routes to project harness initialization. 当需要项目初始化、目标模式、OpenSpec 优先或 Superpowers fallback 的 spec/technical design/plan/evidence、项目 agent 团队、项目 skill、AGENTS.md、交付验证或 Loop Engineering 基座时使用。
| name | knowledge-promotion |
| description | Use when repeated project knowledge, user corrections, quality dissatisfaction, SOP patterns, project skill updates, AGENTS.md updates, templates, or eval cases should be captured and promoted. 当需要沉淀项目经验、踩坑、用户反馈、项目 skill、AGENTS.md、模板、测试夹具或自动迭代 skill 时使用。 |
目标：把开发中的重复解释和质量反馈转成可 review、可验证、可提升的 learning candidate，而不是直接把临时经验写进全局 skill。
learning_candidate.json。AGENTS.md 或项目 skill。python3 skills/knowledge-promotion/scripts/validate_learning_candidate.py learning_candidate.json