- Home
- Skills
- Willoscar
- Research Units Pipeline Skills
- Claim Matrix Rewriter
claim-matrix-rewriter_skill
- Python
109
GitHub Stars
1
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill willoscar/research-units-pipeline-skills --skill claim-matrix-rewriter- SKILL.md2.8 KB
Overview
This skill rewrites outline/claim_evidence_matrix.md as an evidence-driven index that maps subsection claims to concrete evidence packs. It converts evidence_drafts.jsonl into a stable, gate-checked bullet matrix without prose or placeholders. The output preserves legacy markers ("- Claim:", " - Evidence:") so downstream QA and writers can consume a predictable format. It enforces citation keys from citations/ref.bib and flags provisional claims when evidence is title-only or abstract-only.
How this skill works
The script ingests outline/subsection_briefs.jsonl and outline/evidence_drafts.jsonl, resolves citation keys against citations/ref.bib, and emits outline/claim_evidence_matrix.md in a strict bullet format. For each subsection it produces a single "- Claim:" entry followed by two or more " - Evidence:" bullets referencing evidence pack keys and ref.bib citations. It refuses to generate prose, placeholders, or invented facts and marks claims as provisional when full text is missing.
When to use it
- You have completed outline/subsection_briefs.jsonl and outline/evidence_drafts.jsonl and want an evidence-first claim index.
- Before QA or drafting to ensure claims map directly to evidence packs.
- When preparing a claim matrix that downstream checks or scripts will parse programmatically.
- When citations/ref.bib is available and must be the single source of truth for citation keys.
Best practices
- Ensure outline/evidence_drafts.jsonl has blocking_missing resolved before running.
- Verify citations/ref.bib contains every cited key; run citation-verifier if necessary.
- Avoid using abstract-only evidence for definitive language; mark such claims as provisional.
- Keep each subsection to one "- Claim:" and at least two " - Evidence:" bullets.
- Run the provided helper script with --workspace to reproduce the matrix deterministically.
Example use cases
- Generate a claim→evidence index for a literature synthesis before drafting methods and results sections.
- Convert evidence packs from a reading sprint into a navigation layer for reviewers and QA engineers.
- Refresh the claim matrix after updating evidence_drafts.jsonl to keep writing aligned with source materials.
- Detect missing citations quickly by failing the rewrite when keys are absent from citations/ref.bib.
FAQ
The rewrite fails; add the missing key to citations/ref.bib or run citation-verifier before rerunning the script.
Can the skill invent or paraphrase facts not present in evidence packs?
No. The skill prohibits inventing facts and limits claims to what evidence packs and ref.bib support.
How are abstract-only or title-only evidence items handled?
Claims backed only by abstracts or titles must be marked provisional and use cautious language; no strong assertions are allowed.