| name | eval-document |
| description | Evaluate loaded documents against a criteria file, producing a structured findings report. Use when the user asks to "evaluate", "review", or "check" documents against criteria. |
| argument-hint | [criteria-file] |
Evaluate Document
Evaluate loaded documents against a criteria file, producing a structured findings report.
Instructions
-
Gather evidence. Run the eval script to search for relevant clauses per criterion:
uv run --with 'chromadb,sentence-transformers' \
"${CLAUDE_PLUGIN_ROOT}/skills/eval-document/document-eval.py" "$ARGUMENTS"
Default criteria: assets/criteria/general-red-flags.md
-
Assess each criterion. The script outputs evidence grouped by criterion.
For each criterion, read the evidence and apply judgment:
- Rate severity:
CRITICAL, HIGH, MEDIUM, LOW, or CLEAR
- Quote the specific clause text
- Note which source document(s) the finding comes from
- Explain your reasoning
-
Produce the report:
# Document Evaluation Report
**Criteria file:** <path>
**Date:** <date>
**Documents searched:** <count>
**Clauses searched:** <count>
## Summary
| # | Criterion | Severity | Source Files | Sections |
|---|-----------|----------|-------------|----------|
| 1 | ... | CRITICAL | RFP 20-020.pdf | 4.1, 12.12 |
## Detailed Findings
### 1. <Criterion Name>
**Severity:** CRITICAL
**Source:** RFP 20-020 Document.pdf
**Sections:** 4.1, 12.12
**Evidence:** > "quoted clause text..."
**Finding:** <explanation>
Available criteria files
assets/criteria/ip-and-ownership.md — focused IP review (3 criteria)assets/criteria/general-red-flags.md — broad scan (6 criteria)- Custom: any markdown file with
## headings as criteria
Constraints
- Documents must be loaded first
- Always cite section numbers, source documents, and quote clause text as evidence
- Rate conservatively — only use CRITICAL for genuine contradictions or impossibilities
- If a criterion finds no issues, rate it CLEAR and say so explicitly
- Look for cross-document contradictions (e.g., main RFP vs addenda)
