citation-auditor - SKILL.md Agent Skill

name: citation-auditor description: Audit a markdown file by chunking it, extracting claims with structured output, routing each claim to verifier skills, aggregating verdicts, and rendering annotated markdown. argument-hint: "<file.md>" disable-model-invocation: true

Audit the markdown file at $0.

Invocation Contexts

This skill has two roles in this repo:

Standalone /audit (default): user-triggered via /audit <file.md> on an existing markdown file. Uses inline mode — per-claim badges injected into the body plus a per-claim audit report at the end. Does not alter checkpoint.json, verification-audit.json, or any WF1 artifact.
WF1 native verifier backend: Step 3 may reuse the verifier pool under .claude/skills/verifiers/, but it must not use this skill's markdown renderer as a DOCX review output. Native WF1 results must go through citation-checker/scripts/adapt-citation-auditor.py and merge-verification-audits.py so the final artifact remains working/verification-audit.json.

There is no automatic Step 10 in this repo. WF1 remains an 8-step review pipeline.

Procedure

Confirm $0 exists and is a markdown file. If it does not, stop and ask for a valid path.
Run: python -m citation_auditor chunk "$0" --max-tokens 3000
Parse stdout as JSON with the schema { "chunks": [...] }.
For each chunk, extract only factual, citation-bearing claims using structured output with this schema:
- text: string
- sentence_span: { start: integer, end: integer }
- claim_type: factual | citation | quantitative | temporal | other
- suggested_verifier: string | null
Do not extract speculation, forecasts, rumors, advocacy, or soft prediction language such as:
- 전망이다
- 예상된다
- 업계 관계자에 따르면
- 가능성이 있다 Unless the sentence also contains a concrete verifiable factual assertion that stands on its own.
Keep claim offsets chunk-relative. Do not convert them to document offsets yourself.
Route each claim to verifier skills:
- If suggested_verifier is set and exactly matches a loaded verifier skill name, use it.
- Otherwise test the claim text against each verifier skill frontmatter patterns using case-insensitive regex matching and use every match.
- If nothing matches, fall back to general-web.
For each (claim, verifier) pair, use the Task tool to dispatch a subagent that loads that verifier skill and receives the claim JSON.
Require each verifier subagent to return only this JSON: { "label": "...", "rationale": "...", "supporting_urls": ["..."], "authority": 0.0 }
supporting_urls may contain either clickable source URLs or plain-language source references when no stable URL exists. Preserve them verbatim and do not invent clickable URLs for non-linkable sources such as precedent search-result IDs.
Build aggregate input JSON locally. The exact top-level shape is { "verdicts": [ <bundle>, ... ] } where each <bundle> covers one (chunk, claim) pair and holds all verifier candidates for that claim. Do not invent other top-level keys.

Exact schema (copy this structure):
```
{
  "verdicts": [
    {
      "chunk": { "index": 0, "text": "<full chunk text>", "segments": [ { "chunk_start": 0, "chunk_end": 44, "document_start": 0, "document_end": 44 } ] },
      "claim": { "text": "<claim sentence>", "sentence_span": { "start": 64, "end": 268 }, "claim_type": "citation", "suggested_verifier": "us-law" },
      "candidates": [
        {
          "claim": { "text": "<same claim as above>", "sentence_span": { "start": 64, "end": 268 }, "claim_type": "citation", "suggested_verifier": "us-law" },
          "verifier_name": "us-law",
          "authority": 0.9,
          "label": "contradicted",
          "rationale": "<Korean rationale from verifier>",
          "evidence": [ { "url": "https://..." } ]
        }
      ]
    }
  ]
}
```
Rules:
- One bundle per (chunk, claim) pair. If the same claim has two verifiers, put both candidates inside the same bundle's candidates array — do NOT create a second bundle for the same claim.
- The top-level claim inside each bundle is the canonical claim; each candidate's claim must match it verbatim.
- evidence items require a non-empty url string. If the verifier returned supporting_urls: [], emit "evidence": [] — do not emit [{"url": ""}].
- label must be one of verified / contradicted / unknown.
- authority must match the verifier skill's declared authority value.
Write that JSON to a temp file and run: python -m citation_auditor aggregate <tmpfile>
Write the aggregate output to a temp file and run:
- Standalone /audit: python -m citation_auditor render "$0" <aggfile> (inline mode is default).
- Do not use render --mode=append for WF1 DOCX review output.
Return only the final annotated markdown unless the user explicitly asked for intermediate JSON.
If claim extraction validation fails, retry once with a repair prompt. If it still fails, skip that chunk and note it briefly.
If a verifier subagent returns invalid JSON, drop that candidate instead of inventing a verdict.
If a line was skipped because it is a forecast, opinion, rumor, or unattributed speculation, treat that as expected behavior rather than an extraction failure.
If the user asks why a forecast or opinion line was not audited, explain that this plugin audits verifiable factual claims and citations, not predictions or commentary.