argument-selfloop_skill

Overview

This skill enforces argument continuity and premise hygiene for drafted sections by producing a private argument ledger and a premise-consistency report. It runs after prose is allowed (C5) and blocks merge until each section’s paragraph-level argument moves and section-level dependencies are explicit and closed. The ledger and TODO artifacts are intermediate checks and must never be merged into the public draft.

How this skill works

For each target section (start with H3 bodies) it reads the text, labels each paragraph with one or more argument actions, and records concise outputs in a per-section JSONL ledger. It builds a compact argument skeleton that maps what each section consumes and produces, and emits a PASS/FAIL TODO with concrete edits required to close argument gaps. The skill validates the existence and coverage of these artifacts and provides a validator script to block pipeline progress until all invariants pass.

When to use it

• After writer-selfloop PASS and before merging a draft in C5 (PROSE allowed).
• When you want to prevent fluent but hollow prose by enforcing explicit argument moves.
• When sections/*.md exist and you need paragraph-level action labels and section dependency mapping.
• When you suspect premise drift, silent topic switches, or missing local conclusions.
• Do not run if evidence packs are scaffolded/thin—route to evidence-selfloop first.

Best practices

• Label every paragraph with at least one argument action (claim, definition_setup, justification, contrast, boundary_failure, local_conclusion).
• Keep ledger entries short: actions + one-sentence outputs, no prose restatement.
• Treat the ledger as writer-only: never copy ledger content into the public draft.
• Fix only what the self-check flags (bridge sentences, definitions, explicit contrasts, boundaries, or deletions/reorders).
• If a section needs new evidence, stop and route the work to evidence-selfloop.

Example use cases

• Run after finishing an H3 cluster to catch early premise drift before it propagates.
• Validate that a methods section defines protocols used by later evaluations and that definitions are stable.
• Detect and fix silent topic switches between adjacent paragraphs in a results subsection.
• Produce an actionable TODO list describing exactly which paragraphs need new definitions or bridge sentences.
• Block merge when a section’s conclusion depends on unstated premises or shifting metric definitions.