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/kmshihab7878/claude-code-setup --skill operating-framework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
基于 SOC 职业分类
| name | Operating Framework |
| description | Task routing, risk assessment, execution modes, and completion standards for meaningful work. |
Macro-level routing for meaningful work. It classifies lane, risk, mode, required artifacts, evidence, and completion standard. It integrates with existing gates and commands; it does not replace them.
Activate on every meaningful task. Trivial tasks such as a single lookup, file read, or quick answer can bypass the full framework, but safety, verification, and approval gates still apply.
For non-trivial work, route before acting:
TASK: [one-line description]
LANE: [Explore | Specify | Build | Verify | Ship | Recover]
RISK: [T0 Safe | T1 Local | T2 Shared | T3 Critical]
MODE: [A Founder | B Elite Engineering | C Recovery | D Secure Delivery]
ARTIFACTS: [what this task must produce]
EVIDENCE: [what proves completion]
COUNCIL: [Solo | Lead+Reviewer | Full Council]
DONE WHEN: [measurable completion condition]
Use /start-task when an explicit header is useful; otherwise route mentally before acting.
| Lane | Use For | Required Standard |
|---|---|---|
| Explore | Vague request, unknown scope, "what if" | Questions, options, or brief before specification |
| Specify | Behavior change or new capability | Spec, acceptance criteria, and plan before build |
| Build | Approved spec or clear plan | Code, tests, docs as needed |
| Verify | Existing work needs validation | Test results and review findings |
| Ship | Verified work ready for delivery | Commit/PR/deploy checklist and evidence |
| Recover | Bug, incident, regression, failure | Root cause before fix, then regression check |
| Risk | Scope | Behavior |
|---|---|---|
| T0 Safe | Local, read-only, reversible | Proceed freely |
| T1 Local | Local workspace writes | Execute narrowly and verify |
| T2 Shared | Shared systems, remotes, data, config | Pause before shared/irreversible action |
| T3 Critical | Production, financial, security, secrets, irreversible | Full review and explicit user approval |
If unsure, escalate one tier. Production is T2+, financial/trading is T3, and security-sensitive work is T2+.
| Mode | When | Typical Sequence |
|---|---|---|
| A Founder | Greenfield product or feature | brainstorm -> spec -> architecture -> plan -> build -> verify |
| B Elite Engineering | Approved spec and clear implementation | plan -> test -> build -> review -> document |
| C Recovery | Bug, failure, incident, regression | debug -> review -> test -> reflect |
| D Secure Delivery | Security-sensitive, deployment, infra | security audit -> review -> build |
See operating-modes.md for detailed mode playbooks.
Every non-trivial task ends with a completion packet or equivalent final report covering:
See completion-and-precedence.md for the full packet template and skill precedence rules.
If the user explicitly says "just do it", "skip routing", "emergency mode", or "no framework needed", streamline the framework. Never skip:
See escape-hatch-and-integrations.md for bypass logging and integration points.
| Need | Read |
|---|---|
| Session contract, lane definitions, risk tiers, decision tree, artifact matrix, enforcement rules | routing-and-artifacts.md |
| Operating mode detail | operating-modes.md |
| Council compositions and decision format | councils.md |
| Memory promotion and benchmark metrics | memory-and-metrics.md |
| Golden workflows | golden-workflows.md |
| Completion packet and skill precedence | completion-and-precedence.md |
| Escape hatch and integration points | escape-hatch-and-integrations.md |