Input Evidence Cataloger

Use this skill for intake and initial content discovery before storyline or drafting work.

Bundled materials

• references/workflow.md
• references/quality-standard.md
• assets/templates/source-register-template.md
• assets/templates/review-notebook-template.md
• assets/templates/open-questions-template.md
• assets/templates/document-map-template.md

Core outcome

Produce a high-quality evidence catalog with:

• source inventory
• archive or folder document map
• factual statements from each source
• provenance and metadata signals
• supersession and version relationships
• open questions for unresolved facts

Required outputs

1. tracking/source-register.md with one row per source
2. tracking/review-notebook.md with factual statement extraction and confidence tags
3. tracking/open-questions.md with missing provenance or conflicting evidence
4. inputs/processed/document-map.md for zip/folder structure and replacement relationships

Quality gate

Apply references/quality-standard.md before finalizing outputs.

Working materials discipline

• keep extracted statements in tracking/review-notebook.md
• keep provenance and version metadata in tracking/source-register.md
• keep uncertainties in tracking/open-questions.md

Back-iteration loop

1. run initial extraction and cataloging
2. re-check supersession and provenance after full inventory
3. reconcile contradictions and refresh factual statements
4. update outputs and unresolved items before handoff

Parsing rules

• Prefer direct extraction over inference.
• Separate facts from assumptions.
• Mark source confidence and ambiguity explicitly.
• If author/date are unavailable, mark as unknown and record where you looked.
• If a file appears superseded, record both files and the replacement rationale.
• For archives, map internal paths and classify each file role.

Suggested sequence

1. inventory all files under inputs/
2. classify by format and evidence role
3. extract machine-readable text where possible
4. map archive/folder structures
5. generate factual statements per document
6. detect supersession chains and unresolved conflicts
7. publish the four required outputs

Do not

• write recommendations during intake
• hide uncertainty
• collapse multiple source versions into one without traceability
• treat contextual files as controlled evidence unless explicitly confirmed