Doc Index

Document Index Generator

Generate structured YAML index files for project documentation.

Usage

/doc-index

No arguments. Operates on the current project directory.

Output Files

Architecture

The skill operates in two modes based on the project's document structure:

• Flat mode (<50% of files have doc_id frontmatter): Path-based classification. Claude categorizes documents by directory structure and generates index files with basic keyword matching.
• Grouped mode (>50% of files have doc_id frontmatter): Claude-curated. Claude reads key documents, understands domain relationships, and generates index files with semantic understanding.

Instructions

Phase 0: Validate Environment

1. Verify the current directory contains .md files
2. Create docs/.index/ directory if needed: mkdir -p docs/.index
3. If bundles.yaml already exists, note the custom: section for preservation
4. If router.yaml already exists, note the id_patterns: section for preservation

Phase 1: Document Discovery

Run these commands to collect project metadata:

# 1. Find all markdown files
find . -name "*.md" -not -path "*/.git/*" -type f | sort

# 2. Count files with doc_id frontmatter
grep -rl "^doc_id:" --include="*.md" . | wc -l

# 3. Total file count
find . -name "*.md" -not -path "*/.git/*" -type f | wc -l

Treat this as the candidate set, not an unconditional manifest contract. Repository policy may intentionally exclude archival issue drafts, generated fixtures, scratch reports, or other Markdown that is not part of the maintained documentation surface. When excluding candidates, state the excluded categories in the report.

Phase 2: Mode Detection

Calculate: doc_id_count / total_count

• If > 50% → Grouped mode → proceed to Phase 3G
• If ≤ 50% → Flat mode → proceed to Phase 3F

Phase 3F: Flat Mode (Path-Based)

For flat mode projects, generate all 4 files using path-based classification:

1. Classify documents by directory path: rules/ → rule, skills/ → skill, agents/ → agent, reference/ → reference, etc.
2. Group into bundles by classification (core, coding, api, workflow, security, etc.)
3. Build cross-reference graph from markdown links between files
4. Generate keyword-based router from bundle categories
5. Description extraction (#625): for each file's description: field in manifest.yaml, call bash scripts/extract-doc-description.sh <path> rather than taking the first non-empty line directly. The helper skips frontmatter, headings, and pure-HTML structural lines, strips inline tags from mixed prose, and truncates to 200 characters. This avoids the historical bug where READMEs that opened with a centered badge block produced literal <p align="center"> descriptions.

Then proceed to Phase 4 (Write) and Phase 5 (Report).

Phase 3G: Grouped Mode (Claude-Curated)

This is the core of the skill. You read documents and generate index files with domain understanding.

Step 3G.1: Collect Structural Data

Run these to gather raw metadata efficiently:

# All section headings with line numbers
grep -rn '^## \|^### ' docs/ 2>/dev/null | head -800

# All doc_ids
grep -rn '^doc_id:' docs/ 2>/dev/null

# File sizes
find docs/ -name "*.md" -type f -exec wc -c {} + 2>/dev/null | sort -n

# Cross-references (markdown links between docs)
grep -rn '\[.*\]([^)]*\.md[^)]*)' docs/ 2>/dev/null | grep -v '\.index/' | head -1500

Step 3G.2: Read Key Documents

Read these selectively — do NOT read entire large documents:

1. SDS (Software Design Specification): Find via doc_id containing SAD or SDS.

   • Read SI definition sections: find ### .*SI-[A-Z] headings, read each SI's definition (~20-40 lines per SI)
   • Each SI definition contains: safety class, implementation language, SOUP dependencies, SRS allocation, functional description, sub-components

2. SRS (Software Requirements Specification): Find via doc_id containing SRS.

   • Read only ## section headings to understand the category structure (SRS-DATA, SRS-CALC, SRS-SEC, etc.)
   • Note each section's line range

3. Reference documents: For each file in the reference directory:

   • Read the first 5-10 lines (title, doc_id, description) to understand its domain topic
   • Do NOT read full content

4. Screen/Flow documents: Read frontmatter only for hazard_ids, screen_id, flow_id

Step 3G.3: Generate manifest.yaml

Write docs/.index/manifest.yaml with this structure:

# docs/.index/manifest.yaml
# Generated by /doc-index
_meta: {schema: "1.0.0", generated: "YYYY-MM-DD", docs: N, size_mb: X.XX}

core:
  - id: DOC-ID
    file: relative/path.md
    title: "Document Title"
    keywords: [keyword1, keyword2]
    req_count: N        # SRS-xxx count (if any)
    sections:
      - {h: "Heading", l: start, e: end, req: [SRS-CAT]}

# UI Screens (compact)
screens:
  - {id: DOC-ID, file: path.md, scr: SCR-NNN, title: "...", hazards: [H-01]}

# UI Flows (compact)
flows:
  - {id: DOC-ID, file: path.md, flw: FLW-NNN, screens: [SCR-001, SCR-002]}

# Reference Documents (subcategorized)
reference:
  regulatory:
    - {id: DOC-ID, file: path.md, title: "..."}
  rendering:
    - {id: DOC-ID, file: path.md, title: "..."}
  # ... infrastructure, security, clinical, business, standards, general

# Placeholders — stubs
placeholders:
  - {id: DOC-ID, file: path.md}

# Reports — skip unless reviewing
reports:
  - path.md

Classification rules:

• ui/screens/ → screens, ui/flows/ → flows, ui/ other → ui_support
• reference/ → reference (subcategorize by content: regulatory, rendering, infrastructure, security, clinical, business, standards)
• placeholders/ → placeholders, reports/ → reports
• Everything else with doc_id → core

Step 3G.4: Generate bundles.yaml

Write docs/.index/bundles.yaml. This is where domain understanding matters most.

# docs/.index/bundles.yaml
# Generated by /doc-index
_meta: {schema: "1.0.0", generated: "YYYY-MM-DD"}

SI Bundles — For each SI found in the SDS:

Read the SI definition section and determine related files using these signals (in priority order):

1. SDS definition section: Include with precise line range from the ### heading
2. SRS allocation: Parse the SRS 할당 / SRS allocation row in the definition table → find the matching SRS ## sections by category code
3. Explicit links: Follow markdown links in the SI definition to referenced documents
4. Domain relevance: Match reference documents whose title/topic relates to the SI's domain (e.g., SI named "DICOM Gateway" → include documents about DICOM protocol, networking, coordinate transforms)
5. Screen documents: Include screens that implement the SI's SRS categories (use functional categories like DATA, CALC, DISP, REPT — skip cross-cutting categories like SAFE, SEC, PERF that appear in most screens)
6. Core documents: Include IDS/DBS/other core docs that share functional SRS categories with the SI
7. UI support docs: If SI has USAB allocation, include UI support documents (design system, etc.)

Format per SI bundle:

si-xx:
  name: "SI Full Name (SI-XX)"
  files:
    - {file: path.md, lines: "start-end", note: "SI-XX def"}
    - {file: srs-path.md, lines: "start-end", note: "Section Name (SRS-CAT-xxx)"}
    - {file: reference/related-doc.md}

Workflow Bundles — For each FLW document:

• Include the flow document itself
• Include all SCR documents referenced in the flow (scan for SCR-NNN patterns)
• Name: workflow-{flow-name}

Cross-Cutting Bundles:

• security: SRS security section + threat model + auth/encryption/protection reference docs
• safety: SRS safety section + risk management + screens with hazard references
• database: Database schema + data-related reference docs
• api: Interface design spec + API reference docs
• testing: Verification/validation plans + test-related docs
• regulatory-submission: Core regulatory deliverables only (SRS, SDS, SVP, VAL, PRD, SDP, TM, DBS, IDS) — exclude reports, audit files, placeholders

Preserve the existing custom: section from the previous bundles.yaml if it exists.

Step 3G.5: Generate graph.yaml

Write docs/.index/graph.yaml:

# docs/.index/graph.yaml
# Generated by /doc-index
_meta: {schema: "1.0.0", generated: "YYYY-MM-DD"}

cascade — Document-level impact chains. For each document that references another (from cross-reference data):

• Group by source doc_id
• For each target, write a semantic "why" based on the document types:
  • Known pairs: PRD→SRS ("SRS derives from PRD"), SRS→SDS ("Requirement allocation"), SRS→SVP ("Test case mapping"), SRS→TM ("Threats map to SRS-SEC/SAFE"), etc.
  • Other pairs: "Referenced in {section heading}" using the line number from the cross-reference
  • Fallback: "Referenced"

screen_cascade — Screen → flow mappings. For screens appearing in 2+ flows, mark as hub screens.

req_chains — For each SRS category (SRS-DATA, SRS-CALC, etc.):

• Find related SI items (from SDS allocation matrix)
• Find related screens (by SRS category references in screen docs)
• Find related hazards (from screen frontmatter hazard_ids)
• Find related test category (TC-{CAT})
• Find related reference docs

hazard_map — From screen frontmatter hazard_ids: map each H-NN to affected screens.

Step 3G.6: Generate router.yaml

Write docs/.index/router.yaml:

# docs/.index/router.yaml
# Generated by /doc-index
_meta: {schema: "1.0.0", generated: "YYYY-MM-DD"}

id_routes — Auto-discover identifier patterns from documents:

• SRS: compound format SRS-{CAT}-{NNN}, source = SRS document, build section_map from SRS ## headings
• TC: from SVP document
• H: from risk management reference doc
• HUS: from validation plan
• SCR: directory-based ui/screens/SCR-{NNN}-*.md
• FLW: directory-based ui/flows/FLW-{NNN}-*.md
• SI: from SDS document, alpha code format SI-{CODE}
• IF: from SDS document
• T: from threat model document

intent_routes — Keyword → bundle mapping based on which bundles were generated.

skip — [placeholders/*, reports/weekly_report/*, reports/monthly_report/*]

Preserve the existing id_patterns: section if present.

Phase 4: Validate Output

After generating all files:

1. Verify all 4 files exist in docs/.index/
2. Check that manifest covers all discovered documents in the intended index scope, and report any excluded candidate categories
3. Check that each SI bundle has ≥3 files (SDS + SRS + at least 1 reference)
4. Check that cross-cutting bundles have reasonable sizes (5-15 files each)

Phase 5: Report

Present results:

## Document Index Generated

| Metric | Value |
|--------|-------|
| Mode | flat/grouped |
| Total documents | N |
| manifest.yaml | X bytes |
| bundles.yaml | Y bytes |
| graph.yaml | Z bytes |
| router.yaml | W bytes |
| SI bundles | N (grouped only) |
| Workflow bundles | N (grouped only) |

Policies

Command-Specific Rules

How Claude Uses the Index Files