doctree-mcp: Skill for Claude Code

Write Documentation

Create a new markdown document in the knowledge base using the doctree-mcp wiki curation tools.

Requires WIKI_WRITE=1 in the MCP server configuration. If the write tools are unavailable, tell the user to add "WIKI_WRITE": "1" to their doctree env config.

Workflow

Follow these steps in order. Do NOT skip the duplicate check — it prevents content sprawl.

Step 1: Check for duplicates

Before writing anything, check if similar content already exists:

find_similar(content: "<the content you plan to write>")

If matches are found with overlap > 0.35 — UPDATE WORKFLOW:

1. Read the existing doc — call navigate_tree(doc_id, root_node_id) to get the full content
2. Decide what to keep vs. replace — identify sections that overlap with your new content
3. Compose the merged content — write the full updated markdown (don't just append)
4. Overwrite — call write_wiki_entry with the same path and overwrite: true:

write_wiki_entry(
  path: "<same path as existing file>",
  frontmatter: { ... },   ← keep existing frontmatter, update tags if needed
  content: "<full merged content>",
  overwrite: true
)

If overlap is below 0.35 — proceed to Step 2 (scaffold a new entry).

Step 2: Generate a scaffold

Get suggestions for path, frontmatter, and structure:

draft_wiki_entry(
  topic: "$ARGUMENTS",
  raw_content: "<the content to turn into a wiki entry>"
)

Review what comes back:

• Suggested path — adjust if needed (e.g., put runbooks in runbooks/, guides in guides/)
• Inferred frontmatter — type, category, tags derived from similar docs
• Glossary hits — acronyms detected in the content
• Duplicate warnings — similar docs that might overlap

Step 3: Dry-run validation

Validate the entry without writing to disk:

write_wiki_entry(
  path: "<path from step 2, adjusted if needed>",
  frontmatter: {
    "title": "<title>",
    "description": "<one-line summary>",
    "type": "<runbook|guide|reference|tutorial|architecture|adr|...>",
    "category": "<domain grouping>",
    "tags": ["<relevant>", "<terms>"]
  },
  content: "<full markdown content>",
  dry_run: true
)