メニュー
Surface the cognitive traps that corrupt a forecast's reasoning (as distinct from its data or math); use when a defensible-looking forecast may rest on biased judgment.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/grammy-jiang/intelligence-analysis-agent --skill cognitive-bias-and-mindset-controlコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Produce the calibrated numeric probability — a draft Judgment — for a question under uncertainty: outside view / base rate first, Fermi-decompose, fold in the bias/deception review findings, adjust with case specifics, then state an explicit number with confidence and dissent. Invoke when a defensible probability is needed (not a hedge word like 'likely') and being over- or under-confident is costly. This is the sole entry point for producing the draft probability — the pipeline commits it only after the orchestrator's Step 9a audit, Step 10 human gate, and Step 10a lock. Runs as Step 9 of the structured-analysis workflow, which owns the independent reviewer audit (Step 9a), the human gate (Step 10), and the commit (Step 10a): this skill produces the number, it does not audit, approve, or lock it.
Runs an open-source (OSINT) investigation to collect and verify evidence for an analytic question — search, fetch through the egress guard, hash/EXIF for archival, archive each item into the evidence-ledger as an UNTRUSTED ingested proposal, then verify geolocation/chronolocation as candidate annotations (confirmed downstream at the human gate), and hand off to grading — never grading the item here. Invoke when a question needs external open-source collection. Not for classified sources, and not a fact-asserting tool — every result is a candidate for human confirmation.
Grade an evidence item before it drives a judgment: rate source reliability (A–F) and information credibility (1–6) on two independent axes, weight by diagnosticity, seek corroboration, check for deception, and record the absence of expected evidence. Invoke when weighing raw reporting into an analysis. The Step-3 grading method of the structured-analysis workflow; the bias-perception-reviewer and analytic-method-reviewer — and, once its security gate clears, the deception-detection-reviewer — check the result. Not for producing the analytic judgment, critiquing an existing grade, or collecting the raw material.
SOC 職業分類に基づく
SKILL.md を表示中
| name | cognitive-bias-and-mindset-control |
| description | Surface the cognitive traps that corrupt a forecast's reasoning (as distinct from its data or math); use when a defensible-looking forecast may rest on biased judgment. |
| kind | skill |
| status | ready |
| provenance | {"principles":["P001","P008","P016","P017","P019","P020","P027","P033","P034","P041","P042","P045","P057"],"claims":["C00694","C00695","C00299","C00300","C01085","C01086","C00980","C00981","C01380","C01456","C00588","C00685"],"evidence":["E00088","E00089","E00021","E00022","E00215","E00216","E00159","E00160","E00369","E00430","E00055","E00085"],"source_anchors":["d88ef7714f29-c0008","3a2b4f822beb-c0007","e3c7c0b4e46f-c0009","e3c7c0b4e46f-c0003","5da0a790f5ae-c0002","5da0a790f5ae-c0006","d88ef7714f29-c0003","5da0a790f5ae-c0014","5da0a790f5ae-c0003","5da0a790f5ae-c0008","5da0a790f5ae-c0000","5da0a790f5ae-c0012"],"authored_from_digest":"0b7b85c0be3b5aacc5ba345128a154e346cf5bac6d604cd3d3c71fb693224912"} |
Surface and counter the cognitive traps that corrupt a forecast's reasoning — as distinct from its data, its model, or its arithmetic. A forecast can rest on a defensible base rate and a proper scoring rule and still be compromised for purely psychological reasons: a mind-set nobody thought to break, a lesson from history misdiagnosed as stupidity rather than complexity or sample bias, an illusion of validity mistaken for evidence, a hindsight or unpacking error left unmanaged, a recent win or loss anchoring the current call, a small sample over-trusted, or a counterfactual argument shaped more by commitment and dissonance reduction than by evidence. This skill locates those traps in a forecast's reasoning trail and prescribes the specific corrective for each, so the review addresses the thinking that produced the number, not only the number itself.
Work through the reasoning trail in this order; not every step will find a flaw, but every step should be checked before the review is called complete.
Attribute poor learning from history to complexity and sample bias, not stupidity (P001). When a forecaster or decision-maker appears to have drawn a bad lesson from history, do not default to "they weren't smart enough." Look instead for the grounded causes: the underlying subject really is complex, the person only observed a small and skewed slice of past cases, the conditions under which they learned were poor (no clean feedback loop), or a settled view of the past is quietly steering the present read. A related blind spot: because people anchor on dramatic outcomes and have no explicit theory of what would count as "surprising," they systematically under-learn from non-events — the crisis that was quietly defused, the aggression that never came. This applies most when a confident forecast rests on a tidy account of past events, when planning stretches past a reliable horizon, or when a past outcome is being used to set expectations for a new case that only superficially resembles it (P001).
Break a fixed mind-set by thinking backwards (P008). When a mind-set has calcified around one reading of the situation, do not keep arguing forward from "will this happen." Flip the direction instead: assume the surprising, low-probability outcome has already occurred, then work backward to explain how it got there. That single move turns the question from whether into how, and it is the most useful lever exactly where mind-sets are most dangerous to leave unchallenged — the rare, high-consequence scenario nobody wants to take seriously (P008).
Practise active open-mindedness (P016). Check whether the forecaster holds their view as a hypothesis still open to disproof, or as a possession being defended. Active open-mindedness means giving disconfirming evidence and dissenting voices genuine weight, counting a revised opinion as a sign of rigor rather than a defeat, and deliberately seeking out sources that will not simply echo the standing view. The discriminating factor is self-critical habit, not raw intelligence — flag any account whose confidence far outruns its self-scrutiny (P016).
Anticipate the four cognitive obstacles (P017). Before accepting a forecast — or while building the review — check it against four standing obstacles: a pull toward the simplest available story, discomfort with ambiguity and dissonance that gets resolved by premature certainty, a need to see the world as more orderly than it is, and a shaky intuitive grasp of chance. For each, ask whether a deliberate corrective was built into the reasoning, or whether the obstacle was simply left to operate unchecked (P017).
Expose the illusion of validity (P020). Treat a forecaster's confidence, and the coherence of the story behind it, as a feeling produced by how easily the pieces fit together — not as a measure of how accurate the forecast will turn out to be. Coherent, confident impressions have coexisted with near-zero predictive accuracy, and the diagnostic tell is that the confidence typically survives even after the forecaster learns the account has no validity. When a review meets "I'm sure because it all fits," ask for the track record, not more narrative (P020).
Produce a findings list, highest-impact first. Each finding follows a fixed four-part shape:
Never close a review with a bare narrative critique and no correction attached, and never let a single named flaw stand in for the full four-part finding — at minimum, produce one complete finding of this shape before returning control to the caller.
../../references/calibration-forecasting-principles-index.md — full statements, applies-when
contexts, and the claim/evidence/source-anchor trace for every principle cited above.This skill body derives solely from P001, P008, P016, P017, P019, P020, P027, P033, P034, P041, P042, P045, P057; see the frontmatter provenance block above for the associated claims, evidence records, and source anchors.
Manage the offsetting hindsight-and-unpacking errors (P034). Two errors pull in opposite directions, and a review has to manage both, not chase down only the familiar one. Hindsight erases how contingent the past genuinely was, making what happened look inevitable after the fact. Unpacking does the opposite: breaking a future into its named alternative branches makes each one feel more possible simply for having been spelled out. Open-minded, fox-style reasoners are, if anything, the more exposed to the unpacking side — capable of asserting an outcome was inevitable while still granting, in the same breath, that its alternatives were live possibilities, a longer stretch of holding both without resolving the contradiction (P034).
Resist anchoring on the last outcome (P019, P027). A common pattern after a policy fails is to keep the same goal but swing to the opposite tactic. That can be the right call, but the reasoning behind it rarely checks whether the discarded approach would actually have failed again, or whether the new situation genuinely resembles the old one; the error sits at the perceptual stage, where having just been burned primes the forecaster to see many new situations as instances of "the thing that hurt us last time" (P019). The mirror-image trap runs the other way: over-correcting for a costly recent failure can mean excess caution that forfeits a genuinely open opportunity, and the two directions are self-reinforcing — success entrenches whoever advocated the winning policy while failure strengthens the dissenters, so an early run of success can lock in a group too invested to ask what happens if its optimistic assumptions turn out wrong (P027). Ask directly whether the current recommendation is anchored to the last outcome rather than to the facts of the present case.
Distrust small samples — apply the law of small numbers (P057). Even trained researchers over-trust conclusions drawn from small samples and underestimate how much a small sample can wobble by chance alone. When a claim rests on a handful of cases, or a sample size was chosen by habit or gut feel rather than by calculation, flag it — the fix is to size the sample by computing the actual risk of drawing the wrong conclusion, not by intuition or tradition (P057).
Watch motivated counterfactual selection and commitment-driven dismissal (P033, P041). Counterfactual arguments are rarely neutral. The more the reasoner dislikes an actor, the harder they will resist any counterfactual that would credit that actor with a turnaround — dissonance reduction wearing the clothes of analysis. It is most predictable in "hedgehog"-style reasoners who work top-down from a single organizing idea, because their preferred counterfactual can usually be guessed from their ideology before the argument is made (P033). Compounding this, once someone is committed to a generalization, expect them to wave away any counterfactual that would undercut it — even on a fact that looks perfectly neutral, once a rival camp sees advantage in arguing the outcome was easy or hard to have gone otherwise; this shows up most in sophisticated experts, not casual partisans, so credentials are no inoculation against it (P041).
Run a structured post-mortem against the recorded prediction (P042, P045). A post-mortem is only as trustworthy as its record: capture the forecaster's original probability judgment exactly as stated at the time, because hindsight bias and belief-system self-defense reinforce each other and unaided memory reliably drifts to match whatever actually happened (P042). Sequence the post-mortem deliberately — first ask the forecaster to recall their own original prediction unaided (this is where memory drift shows up), then confront them with the actual recorded figure, then probe which self-protective explanation they reach for. Hold open, throughout, that a low-probability event can simply happen — a post-mortem should never treat "keep the number" or "revise the number" as automatically the right answer (P045).