agentleFS
JimmyVGDYJimmyVGDY/codex-long-term-assistant-skills/skills/controlled-evolution-governance/SKILL.md

controlled-evolution-governance

Skill3 starsChanged 2 months agoApache-2.0View on GitHub

仅在跨任务复盘、自观察数据治理、模型成本路由、Reviewer 收益评估、Skill 路由偏差、优化提案生命周期、项目隔离或长期技术助手自身版本治理时使用。普通编码、普通 Review、单次故障排查不要触发。

The file

Files
SKILL.md

受控演进与项目治理

使用边界

  1. 只处理结构化观测事实、统计、证据引用与优化提案,不直接修改其他 Skill、Reviewer、路由策略、业务仓库或生产环境。
  2. 所有 Optimization Proposal 必须保持 execution_authorization=NONE。
  3. 人工 ACCEPT 只表示“允许建立实施任务”,不等于授予文件写入、Git 提交、推送、部署或生产操作权限。
  4. 观察必须按 project_id + repo_fingerprint 双重隔离;任一不一致立即拒绝聚合。
  5. 原始事件先按 event_id 去重,再按 task_id 聚合;同一 Task 不得因生命周期事件数量更多而被重复加权。
  6. 终态结果只使用 PASS/BLOCKED/FAILED/CANCELLED/PARTIAL/UNKNOWN;禁止从通用 status 猜测任务成败。
  7. Reviewer、Explorer、Worker 共用根任务预算;新任务采用脚本控制的 GPT-6 九档默认与有界调整,旧根按冻结策略回放。缺校准数据不妨碍默认委派;自动 xhigh/max/ultra 仍禁止。宿主实际模型身份不读取、不推断,也不参与治理。
  8. Hook 只采集最小结构化元数据,禁止保存原始 Prompt、完整回答、代码正文、Patch、Token、Cookie、API Key 或其他凭据。
  9. health 门禁通过后才运行分析;增量自动化必须按项目显式启用。保持 NO_CHANGE/WAITING_FOR_TASKS/COOLDOWN 安静,仅报告有意义的变化、新候选或需要处理的异常。
  10. 新提案固定可检验假设;实施验证与收益证明分开,收益报告读回时重验引用、窗口、独立样本和质量底线。已确认根因只生成待审回归候选,不自动实施或跨项目晋升。

标准流程

Lifecycle Hooks
    ↓
TaskOutcomeEvent V3
    ↓ event_id 去重
Task 聚合
    ↓ project_id + repo_fingerprint 隔离
Self Observation Snapshot
    ↓
Value / Complexity Assessment
    ↓
Optimization Proposal
    ↓
人工 ACCEPT / REJECT / DEFER
    ↓ ACCEPT 后另建实施 Task
正常 Approval / Git Baseline / Review / Validation
    ↓
关闭 Proposal

详细契约按需读取:

  • references/task-outcome-event-v3.md
  • references/proposal-lifecycle-v6.md

模型与成本原则

主 Agent 保持当前选择。子 Agent 的型号、思考强度和预算采用脚本决策与弹性规则;执行脚本返回的精确参数。

聚合、去重、窗口和成本计算交脚本;跨任务语义冲突保留模型判断和有界调整。缺少统计样本只意味着暂不更新策略参数,不阻止当前默认委派,也不能用更强模型掩盖数据损坏。

DelegationBudget 校准只消费主协调 Agent 已最终化、项目身份完整、批准档位与成本依据明确的样本。离线回放不足最低样本时必须返回“不调整”;任何建议都保持 execution_authorization=NONE,不得直接修改预算或路由。

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.