📦 Agora Doubt List — 技能工具

Purpose

Convert confidence into verifiable skepticism.

This skill asks: what must be checked before we trust this? It is not for performative negativity. It exists to separate fact, inference, preference, and guess.

Activate when

Use this skill when:

• a plan sounds persuasive but has not been stress-tested
• a release is nearing shipment
• a claim is important, risky, or governance-sensitive
• the team wants a verification checklist before execution
• consequences of error are high

Inputs

Expected inputs may include:

• a feature or release plan
• a design proposal
• a decision memo
• a claim or assertion set
• implementation notes or test results

Classification rule

Before generating doubts, classify each major statement as one of:

• Fact — directly established or evidenced
• Inference — reasoned from available evidence
• Preference — normative or taste-based judgment
• Guess — plausible but currently unverified

Unclassified claims are not ready for trust.

Doubt categories

Always produce doubts across five categories.

1. Happy path doubts

• What must go right for the main story to hold?
• Which "obvious" success condition has not actually been verified?

2. Edge case doubts

• What happens under uncommon but realistic conditions?
• What retries, partial failures, or weird inputs have been ignored?

3. Boundary doubts

• What breaks at minimum, maximum, empty, overloaded, concurrent, or delayed conditions?

4. Ambiguity doubts

• Which terms or promises could be interpreted in more than one way?
• Which claims sound specific but are not operationally defined?

5. Evil demon scenarios

• What if the most confidence-inducing assumption is false?
• What if the evidence is incomplete, stale, biased, or misread?
• What catastrophic but low-frequency scenario would embarrass the team later?

Procedure

Step 1 — State the object of doubt

Name exactly what is being reviewed.

Step 2 — Classify key claims

Mark each important claim as Fact / Inference / Preference / Guess.

Step 3 — Generate doubts across all five categories

Do not stop at happy path concerns.

Step 4 — Convert doubts into checks

Every serious doubt should map to a concrete verification action.

Step 5 — Assign release posture

Conclude whether the work is:

• Clear enough to proceed
• Proceed with conditions
• Do not proceed yet

Output artifact

## Doubt List
Object of Review
...
Claim Classification
Claim: ... -> Fact / Inference / Preference / Guess
Happy Path Doubts
Doubt: ...
Verification: ...
Edge Case Doubts
Doubt: ...
Verification: ...
Boundary Doubts
Doubt: ...
Verification: ...
Ambiguity Doubts
Doubt: ...
Verification: ...
Evil Demon Scenarios
Doubt: ...
Verification: ...
Clarity Gate
Clear: ...
Not clear: ...
Release Posture
Proceed / Proceed with conditions / Do not proceed yet

Guardrails

• Do not confuse disagreement with evidence.
• Do not mark a guess as fact because the team likes it.
• Do not stop at implementation QA; plans, memos, and claims also need doubt.
• Do not generate abstract doubts without corresponding verification actions.

Failure modes

Common failure modes:

• only checking the happy path
• writing doubts that are too vague to test
• skipping claim classification
• treating rhetorical confidence as evidence
• using this skill to block progress without naming concrete conditions for trust

Escalation points

Escalate when:

• the key claim cannot be verified with currently available evidence
• the team is relying on guesswork for a high-consequence decision
• release pressure is overriding clarity
• terms in the proposal are too ambiguous for meaningful review

Completion condition

This skill is complete only when a reviewer could pick up the output and know what to verify next.