By name: semi-formal-reasoning description: Applies structured informal-to-formal argument analysis to wiki claims and synthesis drafts. Use when evidence-verifier needs to assess whether stated conclusions follow from cited premises, or when a draft conflates correlation with causation, uses modal hedging incorrectly, or makes unsupported inferential leaps.
Semi-Formal Reasoning
Overview
This skill documents the argument-structure analysis step for the evidence-verifier
persona. "Semi-formal" reasoning means applying logical structure checks without
requiring full first-order logic proofs — the goal is to flag unsupported inferential
moves, scope overreach, and modal errors that a careful reader would catch.
Doc-only workflow — read-only only. No logic/ dir is introduced in MVP.
Any automated flagging tool that writes annotations requires its own explicit
write authorization.
Classification
- Mode: Doc-only workflow — read-only only
- MVP status: Active
- Execution boundary: Read-only argument analysis. Annotation persistence requires separate write authorization.
When to Use
evidence-verifieris reviewing a synthesis draft for logical soundness- A claim states causation when the cited evidence only shows correlation
- Premises are temporally or geographically scoped but the conclusion is not
- A draft uses hedging language ("may", "might") in a way that obscures whether the claim is supported or speculative
- An escalation artifact needs to document the specific inferential flaw rather than just flagging a general "quality concern"
Contract
- Input: a synthesis draft, claim set, or escalation bundle from
evidence-verifierorpolicy-arbiter - Output: a structured reasoning-gap report listing:
- the specific claim or sentence being analyzed
- the premises cited in support
- the inferential move being made (e.g., correlation-to-causation, scope extension, modal drift)
- a recommended repair action (reword, add qualifier, escalate to human)
- Handoff: the reasoning-gap report becomes part of the evidence review bundle
routed to
policy-arbiter; it does not directly modify the draft
Assertions
- This skill is read-only; it does not modify wiki pages or synthesis drafts
- Reasoning gaps are flagged as structured findings, not as editorial changes
- When a gap cannot be resolved by adding a qualifier or citation, the finding
escalates to
policy-arbiterrather than silently approving the draft - Missing or unavailable premises are surfaced as evidence gaps, not as proof by absence
References
AGENTS.mdraw/processed/SPEC.md.github/skills/claim-inventory/SKILL.md.github/skills/verify-citations/SKILL.md.github/skills/enforce-npov/SKILL.md.github/agents/evidence-verifier.md.github/agents/policy-arbiter.md