Skillquality 0.47

recipe-quality-audit

>-

Price
free
Protocol
skill
Verified
no

What it does

recipe-quality-audit

Audit a single NVCA recipe's quality and produce a structured scorecard.

Security model

  • This skill operates on local repository files only — no network access required.
  • Source document downloads (Tier 2 checks) use ensureSourceDocx() which fetches from known template source URLs only.
  • No credentials or API keys are needed.

Usage

Run the audit for a specific recipe:

Audit the recipe: nvca-certificate-of-incorporation

Or audit all recipes:

Audit all NVCA recipes and update the quality tracker

Checks

Tier 1: Structural (no source download needed)

#CheckHow
S1File inventoryDoes recipe have metadata.yaml, replacements.json, clean.json? Optional: computed.json, normalize.json, selections.json
S2Metadata schema validRun existing validateRecipeMetadata() from src/core/metadata.ts
S3Field-to-replacement coverageFor each field in metadata, is there a replacement key referencing {field_name}?
S4Ambiguous keysFlag replacement keys < 8 chars without context qualifier (e.g., [name], [its])
S5Smart quote coverageKeys with apostrophes should have smart-quote variants (or patcher normalizes — check patcher has normalizeQuotes)
S6Source SHA presentsource_sha256 in metadata.yaml
S7Test fixture existsintegration-tests/fixtures/{recipe-id}-*.json exists

Tier 2: Behavioral (requires source download)

#CheckHow
B1Source download + scanCount all \[[_A-Z] prefixed bracket patterns in source (underscore-fill or capitalized placeholders)
B2Replacement coverage ratio(keys with match in source) / (total bracket patterns). Target: >80%
B3Unmatched underscore patterns[___+] patterns in source not in replacements.json
B4Clean effectivenessAfter clean, no footnotes, no "Note to Drafter", no preamble

Tier 3: Fill quality (requires fill run)

#CheckHow
F1Default-only fillFill with defaults, run verifyOutput, count blank placeholders
F2Full-values fillFill with all fields from test fixture, assert all verify checks pass
F3Formatting anomaly countCheck 8 from verifier (single-char underlined runs)
F4Zero-match replacement keysKeys that existed in replacements.json but matched nothing in the source

Output: Quality Scorecard

{
  "recipe_id": "nvca-voting-agreement",
  "maturity": "beta",
  "scores": { "structural": "6/7", "behavioral": "3/4", "fill": "0/4", "total": "9/15" },
  "checks": [
    { "id": "S1", "name": "File inventory", "passed": true, "details": "metadata.yaml, replacements.json, clean.json present" },
    { "id": "S7", "name": "Test fixture exists", "passed": false, "details": "No fixture matching integration-tests/fixtures/nvca-voting-agreement-*.json" }
  ],
  "field_coverage": { "metadata_fields": 14, "replacement_refs": 10, "uncovered": 4 },
  "recommendations": [
    "Add test fixture for fill testing (S7)",
    "Add replacement keys for 4 uncovered fields (S3)"
  ]
}

Maturity Tiers

  • scaffold: metadata-only, can't fill
  • beta: has replacements + clean, score < 11/15 OR no test fixture
  • production: score >= 11/15 AND has test fixture AND has computed.json (if conditional sections exist in source)

Workflow

When running the audit:

  1. Read content/recipes/QUALITY_TRACKER.md for current state
  2. Run Tier 1 checks (always possible)
  3. Run Tier 2 checks if source document is available (use ensureSourceDocx())
  4. Run Tier 3 checks if a test fixture exists
  5. Compute scorecard and maturity tier
  6. Output the scorecard as JSON
  7. Update the quality tracker if requested

Implementation Notes

  • Use validateRecipeMetadata() from src/core/metadata.ts for S2
  • Use ensureSourceDocx() from src/core/recipe/downloader.ts for B1-B4
  • Use runRecipe() from src/core/recipe/index.ts for F1-F4
  • Bracket pattern detection uses \[[_A-Z] prefix to avoid counting citations and legal references
  • Zero-match keys come from PatchResult.zeroMatchKeys returned by the patcher
  • Cross-reference zero-match keys with cleanConfig.removeRanges and cleanConfig.removeParagraphPatterns to suppress expected zero-matches

Capabilities

skillsource-open-agreementsskill-recipe-quality-audittopic-agent-skillstopic-anthropictopic-claudetopic-claude-codetopic-claude-code-clitopic-claude-code-commandstopic-claude-code-plugintopic-claude-code-pluginstopic-claude-code-skillstopic-claude-code-subagentstopic-claude-skillstopic-contract-automation

Install

Quality

0.47/ 1.00

deterministic score 0.47 from registry signals: · indexed on github topic:agent-skills · 31 github stars · SKILL.md body (4,341 chars)

Provenance

Indexed fromgithub
Enriched2026-05-01 06:57:03Z · deterministic:skill-github:v1 · v1
First seen2026-04-18
Last seen2026-05-01

Agent access