gh-slamb2k-mad-skills-dev-flow/skills/cyberarian/SKILL.md

name, description

name

description

cyberarian

The digital librarian for Claude Code projects. Enforces structured document lifecycle management - organizing, indexing, and archiving project documentation automatically. Use when creating, organizing, or managing project documentation. Ensures documents are created in the proper `docs/` directory structure with required metadata, handles temporary documents in system temp directories, maintains an auto-generated index, and performs automatic archiving of old/complete documents. Use for any task involving document creation, organization, or maintenance.

Cyberarian - Document Lifecycle Management

This skill enforces a structured approach to documentation in Claude Code projects, ensuring consistency, discoverability, and automatic maintenance.

Core Principles

1. Structured Organization: All persistent documentation goes in docs/ with semantic categorization
2. No Temporary Docs in docs/: Ephemeral/scratch documents belong in /tmp or system temp, never in docs/
3. Metadata-Driven: YAML frontmatter enables automation and lifecycle management
4. Automatic Maintenance: Indexing and archiving happen automatically, not manually
5. Context Efficiency: Bulk operations delegate to subagents to preserve main context

Context-Efficient Operations

The Problem

Document management operations can produce verbose output that pollutes the main agent's context:

• Validation scripts listing many errors across files
• Index generation scanning dozens of documents
• Archive operations listing all files being moved
• Search results returning many matches

The Solution: Subagent Delegation

Delegate to Task subagent for operations that return verbose output. The subagent absorbs the verbose output in its isolated context and returns a concise summary (<50 tokens).

Delegation Rules

Execute directly (simple, low-output):

• Creating a single document from template
• Reading a specific document's metadata
• Checking if docs/ directory exists

Delegate to Task subagent (complex, verbose):

• Running validation across all documents
• Regenerating the index
• Archiving operations (especially dry-run)
• Searching documents by tag/status/category
• Summarizing INDEX.md contents
• Any operation touching multiple files

Delegation Pattern

When verbose output is expected:

1. Recognize the operation will be verbose
2. Delegate to Task subagent with explicit instructions
3. Subagent executes scripts, absorbs output
4. Subagent parses and returns summary <50 tokens
5. Main agent receives only essential summary

Task subagent prompt format:

Execute document operation and return concise summary:
- Run: [command]
- Parse: Extract [specific data needed]
- Return: [emoji] [state] | [metric] | [next action]
- Limit: <50 tokens

Use agents/doc-librarian-subagent.md patterns for response formatting.

Response Formats

Success: ✓ [result] | [metric] | Next: [action] List: 📋 [N] items: [item1], [item2], ... (+[remainder] more) Error: ❌ [operation] failed | Reason: [brief] | Fix: [action] Warning: ⚠️ [concern] | Impact: [brief] | Consider: [action]

Directory Structure

docs/
├── README.md           # Human-written guide to the structure
├── INDEX.md            # Auto-generated index of all documents
├── ai_docs/           # Reference materials for Claude Code (SDKs, APIs, repo context)
├── specs/             # Feature and migration specifications
├── analysis/          # Investigation outputs (bugs, optimization, cleanup)
├── plans/             # Implementation plans
├── templates/         # Reusable templates
└── archive/           # Historical and completed documents
    ├── specs/
    ├── analysis/
    └── plans/

Workflows

First-Time Setup

When a project doesn't have a docs/ directory:

1. Initialize the structure:

   python scripts/init_docs_structure.py
   

   This creates all directories, README.md, and initial INDEX.md

2. Inform the user about the structure and conventions

Creating a New Document

When asked to create documentation (specs, analysis, plans, etc.):

1. Determine the category:

   • ai_docs: SDKs, API references, repo architecture, coding conventions
   • specs: Feature specifications, migration plans, technical designs
   • analysis: Bug investigations, performance analysis, code audits
   • plans: Implementation plans, rollout strategies, task breakdowns
   • templates: Reusable document templates

2. Use the template:

   cp assets/doc_template.md docs/<category>/<descriptive-name>.md
   

3. Fill in metadata:

   • Set title, category, status, created, last_updated
   • Add relevant tags
   • Start with status: draft

4. Write the content following the document structure

5. Update the index:

   python scripts/index_docs.py
   

File naming convention: Use lowercase with hyphens, descriptive names:

• ✅ oauth2-migration-spec.md
• ✅ auth-performance-analysis.md
• ❌ spec1.md
• ❌ MyDocument.md

Working with Existing Documents

When modifying existing documentation:

1. Update metadata:

   • Set last_updated to current date
   • Update status if lifecycle changes (draft → active → complete)

2. Regenerate index if significant changes:

   python scripts/index_docs.py
   

Creating Temporary/Scratch Documents

When creating ephemeral documents (scratchpads, temporary notes, single-use docs):

NEVER create in docs/ - Use system temp instead:

# Create in /tmp for Linux/macOS
/tmp/scratch-notes.md
/tmp/debug-output.txt

# Let the system clean up temporary files

Why: The docs/ directory is for persistent, managed documentation. Temporary files clutter the structure and interfere with indexing and archiving.

Regular Maintenance

When to run:

• After creating/modifying documents: Update index
• Weekly/monthly: Run archiving to clean up completed work
• Before commits: Validate metadata

Maintenance workflow (delegate to Task subagent for context efficiency):

1. Validate metadata → Delegate to subagent:

   Task: Run python scripts/validate_doc_metadata.py
   Return: ✓ [N] valid | [N] issues: [list top 3] | Next: [action]
   

2. Archive old documents → Delegate to subagent:

   Task: Run python scripts/archive_docs.py --dry-run
   Return: 📦 [N] ready for archive: [list top 3] | Next: Run archive
   
   Task: Run python scripts/archive_docs.py
   Return: ✓ Archived [N] docs | Categories: [list] | Index updated
   

3. Update index → Delegate to subagent:

   Task: Run python scripts/index_docs.py
   Return: ✓ Index updated | [N] documents | Categories: [summary]
   

Why delegate? These operations can scan dozens of files and produce verbose output. Subagent isolation keeps the main context clean for reasoning.

Archiving Documents

Archiving happens automatically based on category-specific rules. See references/archiving-criteria.md for full details.

Quick reference:

• specs/: Auto-archive when status: complete AND >90 days
• analysis/: Auto-archive when status: complete AND >60 days
• plans/: Auto-archive when status: complete AND >30 days
• ai_docs/: Manual archiving only
• templates/: Never auto-archive

To prevent auto-archiving, set in frontmatter:

archivable_after: 2025-12-31

Metadata Requirements

Every document must have YAML frontmatter. See references/metadata-schema.md for complete schema.

Minimal required frontmatter:

---
title: Document Title
category: specs
status: draft
created: 2024-11-16
last_updated: 2024-11-16
tags: []
---

Lifecycle statuses:

• draft → Document being created
• active → Current and relevant
• complete → Work done, kept for reference
• archived → Moved to archive

Reference Files

Load these when needed for detailed guidance:

• references/metadata-schema.md: Complete YAML frontmatter specification
• references/archiving-criteria.md: Detailed archiving rules and philosophy
• agents/doc-librarian-subagent.md: Subagent template for context-efficient operations

Scripts Reference

All scripts accept optional path argument (defaults to current directory):

• scripts/init_docs_structure.py [path] - Initialize docs structure
• scripts/index_docs.py [path] - Regenerate INDEX.md
• scripts/archive_docs.py [path] [--dry-run] - Archive old documents
• scripts/validate_doc_metadata.py [path] - Validate all metadata

Common Patterns

Creating a Specification

# Copy template
cp assets/doc_template.md docs/specs/new-feature-spec.md

# Edit with proper metadata
# category: specs
# status: draft
# tags: [feature-name, relevant-tags]

# Update index
python scripts/index_docs.py

Completing Work

# Update document metadata
# status: draft → active → complete
# last_updated: <current-date>

# After a while, archiving script will auto-archive
python scripts/archive_docs.py

Finding Documents

Delegate searches to subagent for context efficiency:

Task: Summarize docs/INDEX.md
Return: 📊 [N] total docs | Categories: [breakdown] | Recent: [latest doc]

Task: Search docs for tag "performance"
Run: grep -r "tags:.*performance" docs/ --include="*.md" | head -10
Return: 📋 [N] docs match: [path1], [path2], ... | Next: Read [most relevant]

Task: Find all draft documents
Run: grep -r "status: draft" docs/ --include="*.md"
Return: 📋 [N] drafts: [list top 5] | Next: [action]

Direct execution (only for quick checks):

# Check if docs/ exists
ls docs/ 2>/dev/null

Best Practices

1. Always use metadata: Don't skip the frontmatter, it enables automation
2. Keep status current: Update as work progresses (draft → active → complete)
3. Use descriptive names: File names should be clear and searchable
4. Update dates: Set last_updated when making significant changes
5. Run maintenance regularly: Index and archive periodically
6. Temp goes in /tmp: Never create temporary/scratch docs in docs/
7. Validate before committing: Run validate_doc_metadata.py to catch issues
8. Delegate bulk operations: Use Task subagents for validation, indexing, archiving, and search to preserve main context

Error Handling

Document has no frontmatter:

• Add frontmatter using assets/doc_template.md as reference
• Run validate_doc_metadata.py to confirm

Document in wrong category:

• Move file to correct category directory
• Update category field in frontmatter to match
• Regenerate index

Archived document still needed:

• Move from archive/<category>/ back to <category>/
• Update status from archived to active
• Remove archived_date and archive_reason fields
• Regenerate index