Runbook

/runbook — Synthesize docs into actionable runbooks

Turn scattered documentation, config files, and tribal knowledge into a structured runbook that anyone can follow. Produces step-by-step procedures with troubleshooting sections, common gotchas, and recovery steps.

Claude Code Triggers

Invoke this skill when user says:

• "/runbook [topic or path]"
• "Create a runbook for..."
• "Document how to..."
• "Write a troubleshooting guide for..."
• "How do I set up...?" (when the answer should be saved as a runbook)
• "Write onboarding docs for..."

Do NOT invoke when:

• User wants a one-off answer (just answer directly)
• User wants a client proposal (use /client-proposals)
• User wants content strategy docs (use /content-strategy)

Inputs

Process

Step 1: Gather sources

Read all provided source documents. If a folder is given, scan for relevant files:

• .md files (SKILL.md, CLAUDE.md, README)
• .mjs / .js / .py files (read header comments and docstrings)
• .json config files (extract structure and key settings)
• .sh scripts (extract usage patterns and flags)

If only a topic is given (no paths), search the codebase for relevant files:

Glob for: **/*{topic}*
Grep for: relevant terms in.claude/ and project files

Step 2: Identify runbook type

Based on the sources and user intent, classify the runbook:

• Setup — "How to install/configure/initialize X from scratch"
• Troubleshooting — "When X breaks, here's how to diagnose and fix it"
• Onboarding — "New to X? Here's everything you need to know"
• Workflow — "How to do X step-by-step (recurring process)"
• Reference — "Quick reference card for X (flags, configs, common commands)"

Step 3: Extract and organize

From the source material, extract:

1. Prerequisites — What's needed before starting
2. Core procedure — Step-by-step instructions
3. Failure modes — What can go wrong (from code comments, error handling, gotchas in MEMORY.md)
4. Recovery steps — How to fix each failure mode
5. Verification — How to confirm everything worked

Step 4: Write the runbook

Use the output format below. Adjust depth based on audience:

• self — Terse, assumes context, focuses on commands and quick reference
• client — Clear, explains why, includes screenshots/examples
• team — Balanced, includes context but stays focused

Step 5: Save and optionally export

Save the runbook as markdown in the appropriate location:

• Internal runbooks → .claude/runbooks/{topic}.md
• Client runbooks → projects/consulting/{client}/docs/{topic}-runbook.md
• Course runbooks → projects/courses/{course}/docs/{topic}-runbook.md

Optionally export to Google Docs for client delivery:

cd.claude/mcp/gdrive && node create-doc-unified.mjs "/path/to/runbook.md" "Title" --client {slug}

{Title} — Runbook

Type: {setup|troubleshooting|onboarding|workflow|reference} Audience: {self|client|team} Last updated: {date} Source files: {list of files synthesized}

Prerequisites

• {prerequisite 1}
• {prerequisite 2}

Procedure

1. {First major step}

{Explanation if audience is client/team}

{command}

Expected output: {what you should see}

2. {Second major step}

{...}

Troubleshooting

Symptom: {what the user sees}

Cause: {why it happens}

Fix:

{recovery command}

Verify: {how to confirm the fix worked}

Symptom: {another failure mode}

{...}

Common gotchas

• {gotcha 1 — discovered from experience or MEMORY.md}
• {gotcha 2}

Quick reference

Related

• {link to related runbook}
• {link to source skill or MCP}

---

## First Use Cases

Priority runbooks to create when this skill is built:

1. **MCP server troubleshooting** — workspace-mcp OAuth gotchas, Xero token refresh, GDrive script failures (sources: MEMORY.md gotchas, `.claude/mcp/` scripts, `gdrive-protocol.md`)
2. **New client setup** — End-to-end from prospect to active client (sources: `consulting-clients.md`, `/new-client` command, `gdrive-config.json`)
3. **Skill creation** — How to build a new skill from scratch (sources: `_schema/SKILL.template.md`, `_schema/AUTHORING.md`, `skill-catalog`, `skill-reviewer`)
4. **Session recall** — How the recall system works and how to maintain it (sources: `recall/SKILL.md`, `session-indexer.py`, MEMORY.md)

---

## Notes

- Runbooks should be living documents — update them when new gotchas are discovered
- `/session-wrap` can flag potential runbook updates when troubleshooting patterns emerge
- Keep runbooks under 200 lines — if longer, split into multiple focused runbooks
- Use the `self` audience for internal tooling, `client` for handoff docs, `team` for GTM-E School materials