evidence-research

name	evidence-research
description	Use when sources need to be converted into a claim ledger with confidence levels.

evidence-research

Lifecycle stage: RESEARCH

Trigger

Use when external claims, markets, users, or technical options need verification.

When not to use

Do not use when this trigger is absent; choose the command or skill that owns the requested state, artifact, and verification gate.

Inputs

• User request or current artifact.
• Known constraints and context.
• Relevant evidence or source links, if available.
• Current Agent Brain state.

Procedure

1. State the current state and target artifact.
2. Identify missing blockers and ask at most three blocking questions.
3. Separate facts, assumptions, hypotheses, and open questions.
4. Apply the anti-rationalization table below.
5. Produce the required artifact: Research Claim Ledger.
6. Add evidence, risks, decision, and next state.

Anti-Rationalization

Shortcut	Rebuttal
"This is obvious."	Write the assumption and evidence. If you cannot, it is not obvious.
"We can do this later."	If the missing step changes the decision, do it now or state the risk.
"The user wants speed."	Reduce scope; do not skip the quality bar.
"This does not need verification."	Every important claim or behavior needs proof.

Verification

• Required artifact exists and is named.
• Facts, assumptions, and open questions are separated.
• Evidence or evidence gaps are explicit.
• Next state is stated.
• Stop conditions are honored.

Output Artifact

Research Claim Ledger

Use templates/research-claim-ledger.md. The artifact should be concise, auditable, and include evidence, blockers, and next action so another agent can resume.

Failure Modes

• Producing advice instead of an artifact.
• Accepting user assumptions without challenge.
• Skipping evidence because the task feels simple.
• Recommending an agent when a simpler system is enough.
• Hiding risks or open questions.

Example

Trigger: a claim needs source-backed evidence before planning or public copy. Action: split the claim into subclaims, inspect canonical sources, mark confidence, and reject unsupported adoption language. Output artifact: templates/research-claim-ledger.md with blockers and next action. Verification: cite source provenance, checked scope, confidence, and recheck trigger.

A claim says, "Developers already want this workflow." The skill should split the claim into source-backed subclaims, record canonical sources, mark unsupported adoption language as unknown, and produce a research claim ledger with recheck triggers before any roadmap or public copy uses the claim.