用于升级 AI 系统、agent workflow、Codex skill、prompt、memory、RAG、tool routing、schema、eval set 或 feedback loop;也用于把 AI 工作单从指令单升级为意图单,并对研究、检索、测试和 AI 对话做 VOI 决策门审计。需要 Intent Work Order、WOOP 任务准入、决策对象、VOI/EVPI/EVSI、UL(Uncertainty Ladder,不确定性阶梯)、OODA、eval、Human Gate、versioning 与 rollback 的受控演化时使用。Use when controlled AI system evolution or a decision-oriented information audit is needed.
npx skills add https://github.com/DY-2026/GameDesignOS --skill paranoia-ai-system-evolver
> Copyright (c) 2026 Paranoia. Licensed under the MIT License.
把 AI 系统演化当成受控系统设计,而不是神秘的自我改良;把信息获取当成决策投资,而不是越多越好的默认动作。
Intent Work Order 先把“帮我写、优化、分析、出图”的指令单翻译成“要改变什么现实、谁验收、不能牺牲什么、AI 可自治到哪里”的作战意图。
WOOP 定义任务意图、验收结果、失败模式和恢复协议。
Decision Object 定义现在到底要决定什么,以及没有新信息时会做什么。
RJR-AI 定义剩余判断权、授权边界和谁能拍板。
VOI 判断哪些信息、检索、追问、实验或 AI 分支值得付出成本。
Scenario VOI Adapter 定义不同使用场景里什么证据才真的会改变行动。
UL(Uncertainty Ladder,不确定性阶梯)控制下一轮释放多少未知,使失败可归因、复杂度逐步增加并经过迁移验证。
OODA 让 agent 用现实反馈刷新地图。
Evals 决定哪些改动值得留下。
Human Gate 防止一次有用突变污染长期系统。
Rollback 让每次提升都可逆。
VOI 的硬规则:真实、新鲜或结构清晰的信息不一定有价值。只有当合理信号可能改变行动、优先级、资源配置或停止条件时,它才具有当前决策价值。
UL 的硬规则:VOI 选最值得消除的未知,UL 控制暴露剂量。每轮默认只释放一个主要不确定性;若失败无法区分主要解释,必须标记 confounded、恢复支架并退回更受控环境,不能继续叠加永久规则。受控样本通过不等于迁移通过,其他维度通过也不自动提升权限或真实后果。
RJR-AI 的硬规则:AI 可以扩大可能性,Workflow 可以压缩混乱,Eval 可以提供反馈,权限系统可以防止越界,知识库可以积累组织记忆;但高耦合、低可逆、证据不足且必须下注的问题,属于人的剩余判断权,agent 只能辅助论证并进入 Human Gate。
用于改动这些层:
不要用它来合理化失控的模型权重改动、静默长期记忆写入、未经批准的全局 skill 安装,或没有 Human Gate 的生产影响行为。它也不是通用热点总结器;没有决策对象时,只允许有预算的探索或明确的信息消费。
prompt、memory、RAG、tool routing、workflow、eval、schema、docs 或 skill。Intent Work Order:reality_to_change:我要改变什么现实;parent_project_goal:服务哪个更大的项目目标;desired_world_state:完成后外部世界应该变成什么状态;verifier_role 与 first_impression_must_understand:谁验收,第一眼必须看懂什么;must_not_sacrifice、ai_can_freely_change、ai_must_not_touch:不能牺牲、可自由改和不允许碰;decision_principles_if_plan_breaks:原计划不成立时按什么原则改方向;failure_signals_to_check_before_delivery 与 retrospective_contract:交付前自查和复盘沉淀。WOOP Task Card:Wish / Intent Spec:目标、输出物、范围与停止条件;Outcome / Evaluation Rubric:验收标准与决策收益;Obstacle / Failure Pattern:目标漂移、过度信任、上下文污染、工具滥用、FOMO 调研、选项爆炸、虚假确定性等内在失败模式;Plan / If-Then Protocol:触发条件、判断者、恢复动作、重试、交还人或 rollback。Decision Object:current_default_action,即没有新信息时的行动;boundary_status: undefined | far | near | locked。RJR-AI 授权门:residual_judgment,由人选择方向。candidate_information_actions 候选信息行动;posterior_update 与 action_if_seen;model_learning / information_consumption;Scenario VOI Adapter,按具体使用场景定义有效证据:skill_evolution:看真实 trace、行为 eval、负迁移、rollback,而不是一次漂亮案例;game_direction:看玩家承诺、核心循环、题材解释规则、生产风险和最小原型信号;experience_diagnosis:看 evidence_id、issue priority、修复动作和下一轮验证是否改变;source_curation:看材料是否改变入库、分类、沉淀或拒绝,而不是只看内容新鲜;content_decision:看选题、角度、标题承诺、论证主线和发布判断是否改变;platform_fact:看当前一手来源、实际平台状态、兼容策略和时效边界;high_risk_action:看是否降低不可逆错误,并默认进入 Human Gate;ai_branch_management:看分支是否改变下一探针,不能改变行动的分支应归档或关闭。UL (Uncertainty Ladder),为下一轮验证设计可归因的环境:UL-L0 | UL-L1 | UL-L2 | UL-L3 | UL-L4 | UL-L5;released_this_round、held_constant、仍保留的支架和失败后果预算;primary_bottleneck,只针对主瓶颈修改;10. 维护紧凑 OODA 状态:
11. 分离 task OODA 和 meta OODA。任务循环完成当前工作;元循环只提出未来系统可考虑的 candidate 改动。
12. 每个演化改动保持 candidate,直到证据、行为 eval、必要审批和 rollback 都存在。
13. 当目标层是 skill,回放代表性任务,检查是否减少低 VOI 分支、是否保留具体负反馈、是否出现更啰嗦、更慢或误触发的负迁移;同时检查受控通过是否能迁移到陌生样本。
14. 满足任一条件即停止继续获取信息:行动对合理信号已稳健、边际 VOI 不高于边际成本、样本门达到、deadline 到达、剩余不确定性不改变行动,或 Human Gate 已承诺执行。
references/value-of-information-playbook.zh-CN.md;英文:references/value-of-information-playbook.en.md。references/intent-engineering-work-order.zh-CN.md;英文备份:references/intent-engineering-work-order.en.md。workflow-run.governance、shadow/warn/enforce 晋升:references/project-workflow-governance.zh-CN.md;英文备份:references/project-workflow-governance.en.md。references/woop-harness-protocol.zh-CN.md;英文:references/woop-harness-protocol.en.md。references/evolution-loop-playbook.zh-CN.md;英文:references/evolution-loop-playbook.en.md。references/uncertainty-ladder-protocol.zh-CN.md;英文备份:references/uncertainty-ladder-protocol.en.md。references/model-compression-playbook.zh-CN.md;英文:references/model-compression-playbook.en.md。references/eval-versioning-playbook.zh-CN.md;英文:references/eval-versioning-playbook.en.md。templates/intent_work_order.md、templates/intent_work_order.zh-CN.md、templates/intent_work_order.en.md;templates/workflow_governance_review.md、templates/workflow_governance_review.zh-CN.md、templates/workflow_governance_review.en.md;templates/voi_decision_gate.md、templates/voi_decision_gate.zh-CN.md、templates/voi_decision_gate.en.md;templates/ooda_voi_state.md、templates/ooda_voi_state.zh-CN.md、templates/ooda_voi_state.en.md;templates/uncertainty_ladder_state.md、templates/uncertainty_ladder_state.zh-CN.md、templates/uncertainty_ladder_state.en.md;机器对象名为 ul_state;templates/evolution_proposal.md、templates/evolution_proposal.zh-CN.md、templates/evolution_proposal.en.md。evals/voi-decision-gate-cases.md 与 evals/voi-decision-gate-cases.en.md。evals/uncertainty-ladder-cases.md 与 evals/uncertainty-ladder-cases.en.md。执行以下动作前必须询问人:
candidate 提升为当前规则;结束时说明:
candidate;Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.
Intelligently organizes your files and folders across your computer by understanding context, finding duplicates, suggesting better structures, and automating cleanup tasks. Reduces cognitive load and keeps your digital workspace tidy without manual effort.
Generates creative domain name ideas for your project and checks availability across multiple TLDs (.com, .io, .dev, .ai, etc.). Saves hours of brainstorming and manual checking.
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.
Implements Manus-style file-based planning for complex tasks. Creates task_plan.md, findings.md, and progress.md. Use when starting complex multi-step tasks, research projects, or any task requiring >5 tool calls.
Creative research ideation and exploration. Use for open-ended brainstorming sessions, exploring interdisciplinary connections, challenging assumptions, or identifying research gaps. Best for early-stage research planning when you do not have specific observations yet. For formulating testable hypotheses from data use hypothesis-generation.
Comprehensive GitHub project management with swarm-coordinated issue tracking, project board automation, and sprint planning
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
Take dy-2026/paranoia-ai-system-evolver from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.