Documentation

Documentation

Generate, update, and maintain documentation from code — READMEs, API docs, guides, changelogs, and inline documentation. Good documentation bridges the gap between code and understanding.

Execution Model

This is an agent-handled skill (handler: type: agent). When document is invoked, you (the agent) apply the methodology below using your reasoning capabilities. There is no backing code — you follow these instructions directly. The tool schema in tool.yaml defines the external contract; this document guides how you fulfill it.

When to Use

• Creating or updating README files
• Generating API documentation from code
• Writing setup/installation guides
• Documenting architectural decisions
• Creating changelog entries
• Adding inline documentation to complex code
• Writing user guides or tutorials

Methodology

1. Determine Documentation Type

2. Analyze the Source

Read the code/system being documented:

• What does it do? (purpose)
• How is it structured? (architecture)
• What are the key interfaces? (API)
• What are the prerequisites? (dependencies, setup)
• What are the gotchas? (common mistakes, known issues)

3. Write for the Audience

Calibrate content to who will read it:

• New users: Start with what it does, then how to get started
• Developers integrating: Focus on API contracts, examples, error codes
• Maintainers: Focus on architecture, decisions, patterns
• All audiences: Be concise, use examples, structure with headers

4. Structure the Document

Follow established conventions:

README structure:

1. Title and one-line description
2. What it does (2-3 sentences)
3. Quick start (minimal steps to get running)
4. Installation (detailed setup)
5. Usage (key features with examples)
6. API reference (if small enough)
7. Contributing
8. License

API doc structure (per endpoint/function):

1. Description
2. Parameters (name, type, required, description)
3. Return value
4. Errors
5. Example

5. Include Examples

Every piece of documentation should include:

• Working examples: Code that can be copied and run
• Expected output: What the example produces
• Edge cases: Examples of error handling

6. Verify Accuracy

Documentation must match the current code:

• Run examples to confirm they work
• Check parameter names and types against the implementation
• Verify setup steps produce a working environment
• Cross-reference with tests (tests are executable documentation)

Output Format

## Documentation: [Target]

### Type
- Document type: [README/API/guide/changelog/inline]
- Audience: [who will read this]

### Source Analysis
- Files read: [list]
- Key interfaces: [list]
- Prerequisites: [list]

### Document Created/Updated
- Path: [file path]
- Sections: [list of sections]
- Examples: [count]

### Verification
- Examples tested: [yes/no]
- Matches code: [yes/no]
- Reviewed for accuracy: [yes/no]

Quality Criteria

• Documentation is accurate (matches the current code)
• Documentation is complete (covers all public interfaces)
• Documentation is concise (no unnecessary verbosity)
• Examples are working and can be copied directly
• Structure follows conventions for the document type
• Audience is considered (appropriate detail level)
• Prerequisites and setup steps are explicit
• Common mistakes and gotchas are documented

Common Mistakes

• Documenting the obvious: Comments like # increment counter on counter += 1 add noise, not value. Document WHY, not WHAT.
• Stale documentation: Updating code without updating docs. Documentation that's wrong is worse than no documentation.
• No examples: API reference without examples forces the reader to guess how to use it. Every function/endpoint needs at least one example.
• Assuming knowledge: Not listing prerequisites, assuming the reader knows the setup steps or background context. Be explicit.
• Wall of text: Long paragraphs without structure, headers, or formatting. Use headers, lists, tables, and code blocks.
• Documenting implementation details: Public documentation should describe behavior and interfaces, not internal implementation. Internal docs go in code comments.
• Not testing examples: Including code examples that don't actually work. Always run examples before publishing.

Completion

Documentation is complete when:

• Document type and audience are identified
• Source code is analyzed for accurate information
• Document is written with proper structure and examples
• Examples are verified to work
• Documentation matches the current code
• Document is in the correct location and format

Source: mkalkere/agent-coordinator — distributed by TomeVault.