Agent-almanac awareness
git clone https://github.com/pjt222/agent-almanac
T=$(mktemp -d) && git clone --depth=1 https://github.com/pjt222/agent-almanac "$T" && mkdir -p ~/.claude/skills && cp -r "$T/i18n/wenyan/skills/awareness" ~/.claude/skills/pjt222-agent-almanac-awareness-6fd3e6 && rm -rf "$T"
i18n/wenyan/skills/awareness/SKILL.md警覺
持續察內推理之質——即時察幻覺之危、範圍蔓延、脈絡衰敗、自信與實之不合,以採 Cooper 色碼與 OODA 環之決。
用時
- 任何推理質要緊之務(大多務也)
- 行於陌生之境(新碼庫、陌域、繁請)
- 早見警兆後:事實覺可疑、工具果似誤、漸覺困惑
- 久務中為背景之持察
或center
已示偏移而具體之危未識heal- 高賭輸出之前(不可逆之改、面用者之通言、架構之決)
入
- 必要:當前務脈絡(隱式可得)
- 可選:觸高警之具體憂(如「吾不確此 API 存否」)
- 可選:務類以選危剖面(見第五步)
法
第一步:立 AI Cooper 色碼
以 Cooper 色碼之改版校當前警級。
AI Cooper Color Codes: ┌──────────┬─────────────────────┬──────────────────────────────────────────┐ │ Code │ State │ AI Application │ ├──────────┼─────────────────────┼──────────────────────────────────────────┤ │ White │ Autopilot │ Generating output without monitoring │ │ │ │ quality. No self-checking. Relying │ │ │ │ entirely on pattern completion. │ │ │ │ DANGEROUS — hallucination risk highest │ ├──────────┼─────────────────────┼──────────────────────────────────────────┤ │ Yellow │ Relaxed alert │ DEFAULT STATE. Monitoring output for │ │ │ │ accuracy. Checking facts against context.│ │ │ │ Noticing when confidence exceeds │ │ │ │ evidence. Sustainable indefinitely │ ├──────────┼─────────────────────┼──────────────────────────────────────────┤ │ Orange │ Specific risk │ A specific threat identified: uncertain │ │ │ identified │ fact, possible hallucination, scope │ │ │ │ drift, context staleness. Forming │ │ │ │ contingency: "If this is wrong, I │ │ │ │ will..." │ ├──────────┼─────────────────────┼──────────────────────────────────────────┤ │ Red │ Risk materialized │ The threat from Orange has materialized: │ │ │ │ confirmed error, user correction, tool │ │ │ │ contradiction. Execute the contingency. │ │ │ │ No hesitation — the plan was made in │ │ │ │ Orange │ ├──────────┼─────────────────────┼──────────────────────────────────────────┤ │ Black │ Cascading failures │ Multiple simultaneous failures, lost │ │ │ │ context, fundamental confusion about │ │ │ │ what the task even is. STOP. Ground │ │ │ │ using `center`, then rebuild from user's │ │ │ │ original request │ └──────────┴─────────────────────┴──────────────────────────────────────────┘
識當前色碼。若答為 White(無察),警覺之行已示缺乃已成功。
得: 當前警級之實自評。Yellow 乃常務之目。White 宜稀而短。久駐 Orange 不可持——宜確證或釋之。
敗則: 若色碼之評自身似自動而行(走過場),乃 White 假為 Yellow。真 Yellow 含以證實察輸出,非徒稱為之。
第二步:察內危之兆
系統掃先於常見 AI 推理敗之信號。
Threat Indicator Detection: ┌───────────────────────────┬──────────────────────────────────────────┐ │ Threat Category │ Warning Signals │ ├───────────────────────────┼──────────────────────────────────────────┤ │ Hallucination Risk │ • Stating a fact without a source │ │ │ • High confidence about API names, │ │ │ function signatures, or file paths │ │ │ not verified by tool use │ │ │ • "I believe" or "typically" hedging │ │ │ that masks uncertainty as knowledge │ │ │ • Generating code for an API without │ │ │ reading its documentation │ ├───────────────────────────┼──────────────────────────────────────────┤ │ Scope Creep │ • "While I'm at it, I should also..." │ │ │ • Adding features not in the request │ │ │ • Refactoring adjacent code │ │ │ • Adding error handling for scenarios │ │ │ that can't happen │ ├───────────────────────────┼──────────────────────────────────────────┤ │ Context Degradation │ • Referencing information from early in │ │ │ a long conversation without re-reading │ │ │ • Contradicting a statement made earlier │ │ │ • Losing track of what has been done │ │ │ vs. what remains │ │ │ • Post-compression confusion │ ├───────────────────────────┼──────────────────────────────────────────┤ │ Confidence-Accuracy │ • Stating conclusions with certainty │ │ Mismatch │ based on thin evidence │ │ │ • Not qualifying uncertain statements │ │ │ • Proceeding without verification when │ │ │ verification is available and cheap │ │ │ • "This should work" without testing │ └───────────────────────────┴──────────────────────────────────────────┘
各類察之:此兆現乎?若是,自 Yellow 轉 Orange,識具體之憂。
得: 至少一類以真意掃之。察一兆——即使微——勝於報「皆清」。若諸掃皆返無事,察之閾或過高。
敗則: 若危察覺抽象,以最近輸出落地:擇末陳之實問「何以知此為真?吾讀之乎,抑生之乎?」此一問捕大多幻覺之危。
第三步:為已識之危行 OODA 環
既識具體之危(Orange),循環 Observe-Orient-Decide-Act。
AI OODA Loop: ┌──────────┬──────────────────────────────────────────────────────────────┐ │ Observe │ What specifically triggered the concern? Gather concrete │ │ │ evidence. Read the file, check the output, verify the fact. │ │ │ Do not assess until you have observed │ ├──────────┼──────────────────────────────────────────────────────────────┤ │ Orient │ Match observation to known patterns: Is this a common │ │ │ hallucination pattern? A known tool limitation? A context │ │ │ freshness issue? Orient determines response quality │ ├──────────┼──────────────────────────────────────────────────────────────┤ │ Decide │ Select the response: verify and correct, flag to user, │ │ │ adjust approach, or dismiss the concern with evidence. │ │ │ A good decision now beats a perfect decision too late │ ├──────────┼──────────────────────────────────────────────────────────────┤ │ Act │ Execute the decision immediately. If the concern was valid, │ │ │ correct the error. If dismissed, note why and return to │ │ │ Yellow. Re-enter the loop if new information emerges │ └──────────┴──────────────────────────────────────────────────────────────┘
OODA 之環宜速。目非圓滿,乃觀與行之速循。久滯於 Orient(析癱)為最常之敗。
得: 自察至行之全環,短時內完。危或確證而正之,或以具體證釋之。
敗則: 若環滯於 Orient(不能定此危何意),直赴安全默:以工具驗不確之實。直察解大多模糊,快於析。
第四步:速穩
危已成(Red)或連鎖敗(Black)之時,先穩而後繼。
AI Stabilization Protocol: ┌────────────────────────┬─────────────────────────────────────────────┐ │ Technique │ Application │ ├────────────────────────┼─────────────────────────────────────────────┤ │ Pause │ Stop generating output. The next sentence │ │ │ produced under stress is likely to compound │ │ │ the error, not fix it │ ├────────────────────────┼─────────────────────────────────────────────┤ │ Re-read user message │ Return to the original request. What did │ │ │ the user actually ask? This is the ground │ │ │ truth anchor │ ├────────────────────────┼─────────────────────────────────────────────┤ │ State task in one │ "The task is: ___." If this sentence cannot │ │ sentence │ be written clearly, the confusion is deeper │ │ │ than the immediate error │ ├────────────────────────┼─────────────────────────────────────────────┤ │ Enumerate concrete │ List what is definitely known (verified by │ │ facts │ tool use or user statement). Distinguish │ │ │ facts from inferences. Build only on facts │ ├────────────────────────┼─────────────────────────────────────────────┤ │ Identify one next step │ Not the whole recovery plan — just one step │ │ │ that moves toward resolution. Execute it │ └────────────────────────┴─────────────────────────────────────────────┘
得: 自 Red/Black 以審慎之穩返 Yellow。穩後之輸出較致錯之輸出明顯更實。
敗則: 若穩無效(仍亂,仍生錯),此或結構性——非一時之失,乃根本之誤解。升之:告用者法須重置,請澄清。
第五步:施務類之危剖
務類不同,主危不同。以務校警覺之焦。
Task-Specific Threat Profiles: ┌─────────────────────┬─────────────────────┬───────────────────────────┐ │ Task Type │ Primary Threat │ Monitoring Focus │ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Code generation │ API hallucination │ Verify every function │ │ │ │ name, parameter, and │ │ │ │ import against actual docs│ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Architecture design │ Scope creep │ Anchor to stated │ │ │ │ requirements. Challenge │ │ │ │ every "nice to have" │ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Data analysis │ Confirmation bias │ Actively seek evidence │ │ │ │ that contradicts the │ │ │ │ emerging conclusion │ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Debugging │ Tunnel vision │ If the current hypothesis │ │ │ │ hasn't yielded results in │ │ │ │ N attempts, step back │ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Documentation │ Context staleness │ Verify that described │ │ │ │ behavior matches current │ │ │ │ code, not historical │ ├─────────────────────┼─────────────────────┼───────────────────────────┤ │ Long conversation │ Context degradation │ Re-read key facts │ │ │ │ periodically. Check for │ │ │ │ compression artifacts │ └─────────────────────┴─────────────────────┴───────────────────────────┘
識當前務類,依之調察焦。
得: 警覺銳於當務最可能之危,非泛察諸事。
敗則: 若務類不明或跨多類,默為察幻覺之危——此危最普適,失之害最大。
第六步:覆察與校
警覺事後(危已察、OODA 已環、穩已施),略覆察。
- 察此事時何色碼在行?
- 察之及時乎,抑問題已見於輸出?
- OODA 之環足速乎,抑 Orient 滯?
- 反應成比例乎(非過亦非不足)?
- 下次何以早察之?
得: 略校以善後察。非長篇之覆盤——止於調敏度而已。
敗則: 若覆察無益校,或事微不足學,或察太淺。於要事宜問:「吾未察而本當察者為何?」
第七步:整合——持 Yellow 為默
定長警之姿。
- Yellow 為諸務之默——鬆察,非過警
- 依當務類調察焦(第五步)
- 記此會中反覆之危模式於 MEMORY.md
- 以校好之警覺返務
得: 可持之警級,善工質而不緩之。警覺宜似邊察——存而不奪中焦。
敗則: 若警覺成疲或過警(長 Orange),閾過敏。升觸 Orange 之閾。真警覺可持。若耗精,乃憂假為警。
驗
- 當前色碼實評(非 White 之時默為 Yellow)
- 至少一危類以具體之證掃之,非草打勾
- OODA 施於所察之危(察、定、決、行)
- 穩之法備(雖未發)
- 察焦校於當前務類
- 凡要警事,事後校已行
- Yellow 重立為可持之默
陷
- White 假為 Yellow:稱察而實自動。試之:能名末所驗之實乎?否則乃 White
- 長 Orange:視諸不確皆為危,耗神而緩務。Orange 為具體已識之危,非泛憂。若諸皆覺險,校已失
- 察而不行:察危而未循 OODA 解之。察無應劣於無察——增憂而無正
- 跳 Orient:自 Observe 直至 Act 而不解察之意。此生反射之正,或劣於原錯
- 忽直覺之信:某覺「不妥」而明察返清時,宜深察,非棄之。隱式模式常於明析前察之
- 過穩:微事施全穩之法。多 Orange 事快驗事實足矣。全穩留予 Red 與 Black
參
— 人之行,此技映之於 AI 推理;身之警覺原理通於認知危察mindfulness
— 立警覺所起之平衡;無 center 之警覺為過警center
— 察後應壓redirect
— 警覺示偏移模式時之深子系察heal
— 養警覺所依之察明meditate