Docs Create

docs-create: External Docs Skill Generator

Creates epic-specific skill for loading external documentation into session context.

Problem Solved

When working on epics that require external docs (Beads, Claude Code, Railway, etc.), you need:

• Same 10-20 docs available across multiple sessions
• Auto-activation when relevant topics come up
• No re-pasting URLs every session
• Easy to add/remove/refresh docs

Solution

Creates .claude/skills/docs-{epic-id}/SKILL.md with:

• 3-6 line summaries per doc (in skill description)
• Full docs cached to Serena
• Auto-activation on relevant topics
• Commands: list, show, add, remove, refresh, archive

Usage

/docs-create bd-xyz https://url1 https://url2 https://url3 ...

Arguments:

• epic-id: Beads issue ID (e.g., bd-xyz)
• urls: Space-separated list of URLs to cache

Example:

/docs-create bd-abc \
  https://github.com/steveyegge/beads/blob/main/AGENTS.md \
  https://docs.claude.com/claude-code/skills \
  https://docs.railway.app/reference/variables

Workflow

Phase 1: Fetch & Cache

For each URL:

1. Download content (WebFetch for web, raw GitHub URLs for .md)
2. Extract meaningful slug from URL
3. Cache to .serena/memories/external_{epic}/doc{N}_{slug}.md
4. Track URL → doc mapping

Storage structure:

.serena/memories/external_bd-xyz/
  ├── doc1_beads_agents.md
  ├── doc2_claude_skills.md
  ├── doc3_railway_variables.md
  └── manifest.json  # URL → doc mapping

Phase 2: Batch Summarization (Built-in)

Read all cached docs together and generate summaries.

Process:

1. Read all N docs from .serena/memories/external_{epic}/
2. Analyze together to identify:
   • Key concepts per doc
   • Overlapping topics (reduce redundancy)
   • Cross-references between docs
   • Trigger keywords for auto-activation
3. Generate 3-6 line summary per doc
4. Extract combined topic keywords

Example output:

Doc 1: Beads AGENTS.md
Summary: Beads workflow for AI agents: Issue-First development pattern, epic→feature→task hierarchy, dependency types (blocks, parent-child, discovered-from). Core commands: bd create/update/close, bd ready, bd sync. Multi-developer git integration via .beads/issues.jsonl.
Topics: beads, workflow, issue-first, dependencies

Doc 2: Claude Code Skills
Summary: Skill structure: SKILL.md frontmatter (name, description, allowed-tools), auto-activation via semantic description matching. Tool restrictions, skill chaining, error handling patterns. Integrates with Beads workflow (see Doc 1).
Topics: claude-code, skills, auto-activation, tools

Doc 3: Railway Variables
Summary: Environment variable management: service-level vs shared variables, reference syntax ${{SERVICE.VAR}}, Railway CLI commands, secrets handling. Auto-redeploy on variable changes. Used for deployment config (complements Claude Code in Doc 2).
Topics: railway, environment, deployment, config

Combined topics: beads, workflow, issue-first, claude-code, skills, railway, deployment, environment

Why batch processing:

• See all docs together (identify overlaps)
• Reduce redundant summaries (cross-reference instead)
• Create cohesive set that works together

Phase 3: Generate Epic-Specific Skill

Create .claude/skills/docs-{epic}/SKILL.md:

---
name: docs-{epic}
description: |
  Documentation context for {epic}: {title}

  Auto-activates when discussing: {combined_topics}

  Summaries ({N} docs):

  1. **{doc1_name}** - {doc1_summary}

  2. **{doc2_name}** - {doc2_summary}

  ... [{N} total]

allowed-tools:
  - Read
  - mcp__serena__read_memory
  - Bash
  - Write
  - WebFetch
  - Task
---

# Documentation Context: {epic}

This skill provides quick access to {N} external docs.

## Cached Docs

{manifest.json content - URL → doc mapping}

## Commands

Run `/docs-{epic} <command>` where command is:

- `list` - Show all doc summaries
- `show <N>` - Read full doc (e.g., `/docs-{epic} show 3`)
- `add <url>` - Add new doc to collection
- `remove <N>` - Remove doc from collection
- `refresh` - Re-download all URLs (get latest)
- `archive` - Move skill to archive (cleanup)

[Command implementations...]

Phase 4: Confirmation

✅ Created .claude/skills/docs-{epic}/
✅ Cached {N} docs to .serena/memories/external_{epic}/
✅ Generated summaries via documentation-engineer
✅ Skill auto-activates on: {topics}

Next:
- Skill is active (will auto-trigger on relevant topics)
- Manage with: /docs-{epic} list|show|add|remove|refresh
- When done: /docs-{epic} archive

Agent Implementation (When Invoked)

When user runs /docs-create bd-xyz <url1> <url2> ..., follow these steps:

Step 1: Parse & Validate

Input: epic_id, urls[]
Validate: epic exists in Beads (bd show {epic_id})
Get: epic title for context

Step 2: Fetch & Cache All Docs

For each URL:

1. Use WebFetch to get content (handles both .md and HTML)
2. Extract slug from URL for filename
3. Write to .serena/memories/external_{epic}/doc{N}_{slug}.md
4. Track in manifest.json

// .serena/memories/external_{epic}/manifest.json
{
  "epic_id": "bd-xyz",
  "epic_title": "Authentication Epic",
  "created": "2025-01-15T14:30:00Z",
  "docs": [
    {"num": 1, "url": "https://...", "file": "doc1_beads_agents.md", "slug": "beads_agents"},
    {"num": 2, "url": "https://...", "file": "doc2_claude_skills.md", "slug": "claude_skills"}
  ]
}

Step 3: Read All Docs & Generate Summaries

Read all cached docs together:

For doc in .serena/memories/external_{epic}/*.md:
  Read full content
  Note key concepts
  Identify overlap with other docs

Generate summaries (3-6 lines each):

• Focus on unique value of each doc
• Cross-reference related docs (e.g., "see Doc 1 for...")
• Extract topic keywords
• Ensure non-redundant (if two docs cover same topic, divide clearly)

Output format:

summaries = [
  {
    num: 1,
    name: "Beads AGENTS.md",
    summary: "...",
    topics: ["beads", "workflow", ...]
  },
  ...
]
combined_topics = [all unique topics]

Step 4: Generate docs-{epic} Skill

Create .claude/skills/docs-{epic}/SKILL.md with:

• Frontmatter: name, description (with summaries), allowed-tools
• Commands: list, show, add, remove, refresh, archive
• Manifest reference: Link to cached docs

See "Generated Skill Template" section below for full template.

Step 5: Confirm

✅ Created .claude/skills/docs-{epic}/
✅ Cached {N} docs to .serena/memories/external_{epic}/
✅ Generated {N} summaries
✅ Skill active - auto-activates on: {topics}

Test: Try asking about one of the topics to verify auto-activation
Manage: /docs-{epic} list|show|add|remove|refresh|archive

Generated Skill Commands

Each generated docs-{epic} skill includes these commands:

list

Shows all cached doc summaries.

show

Reads full doc from Serena cache.

DOC_NUM=$1
EPIC_ID=$(basename $(dirname "$0") | sed 's/docs-//')
mcp__serena__read_memory(
  memory_file_name="external_${EPIC_ID}/doc${DOC_NUM}_*.md"
)

add

Adds new doc to collection, regenerates summaries.

remove

Removes doc from collection, regenerates summaries.

refresh

Re-downloads all URLs (gets latest versions).

archive

Moves skill to .claude/skills/archive/ for cleanup.

Best Practices

When to use:

• Starting work on epic with 5+ external docs
• Docs needed across multiple sessions
• Want auto-activation on relevant topics

When NOT to use:

• One-off doc reference (just use WebFetch)
• Docs change frequently (manual refresh overhead)
• Simple features (<5 docs, 1-2 sessions)

Maintenance:

• Run /docs-{epic} refresh weekly if docs update frequently
• Remove unused docs with /docs-{epic} remove N
• Archive when epic complete: /docs-{epic} archive

Example Workflow

# Day 1: Start authentication epic
User: "work on authentication"
Agent: Creates bd-abc

# Day 1: Load external docs
User: "/docs-create bd-abc https://beads.md https://railway.md https://clerk.md"
Agent: [Fetches, summarizes, creates skill]
       ✅ docs-bd-abc active

# Day 2: Work on spec (docs auto-activate)
User: "how do I create epic dependencies in Beads?"
Agent: [docs-bd-abc auto-activates]
       [Reads Beads summary, decides to show full doc]
       /docs-bd-abc show 1
       [Answers with full context]

# Day 10: Need more docs
User: "/docs-bd-abc add https://supabase.com/docs/auth/rls"
Agent: [Fetches, regenerates summaries]
       ✅ Now 4 docs

# Week 4: Complete epic
User: "finish bd-abc"
Agent: [finish-feature skill]
       "/docs-bd-abc archive"
       ✅ Archived

Storage & Cleanup

Active:

.claude/skills/docs-bd-xyz/SKILL.md
.serena/memories/external_bd-xyz/
  ├── manifest.json
  ├── doc1_beads_agents.md
  └── doc2_claude_skills.md

Archived:

.claude/skills/archive/docs-bd-xyz/SKILL.md
.serena/memories/archive/external_bd-xyz/

Restore:

mv .claude/skills/archive/docs-bd-xyz .claude/skills/
mv .serena/memories/archive/external_bd-xyz .serena/memories/
/docs-bd-xyz refresh  # Get latest versions

Generated Skill Template

Template for .claude/skills/docs-{epic}/SKILL.md:

---
name: docs-{epic}
description: |
  Documentation context for {epic}: {epic_title}

  Auto-activates when discussing: {combined_topics}

  Summaries ({N} docs):

  1. **{doc1_name}** - {doc1_summary}

  2. **{doc2_name}** - {doc2_summary}

  ... [{N} total]

allowed-tools:
  - Read
  - mcp__serena__read_memory
  - Write
  - WebFetch
---

# Documentation Context: {epic}

{epic_title}

**Cached docs:** {N}
**Last updated:** {date}
**Location:** `.serena/memories/external_{epic}/`

## Quick Reference

{summary_list}

## Commands

Run `/docs-{epic} <command>`:

- `list` - Show all summaries
- `show <N>` - Read full doc
- `add <url>` - Add new doc
- `remove <N>` - Remove doc
- `refresh` - Re-download all
- `archive` - Move to archive

## Cached Docs

{manifest_content}

---

When invoked with command:
- Parse command (list/show/add/remove/refresh/archive)
- Execute appropriate action
- Use mcp__serena__read_memory for showing full docs
- Use WebFetch for adding/refreshing

Related

• issue-first skill: Suggests /docs-create for epics
• finish-feature skill: Auto-archives docs when epic complete