documentation

name	documentation
description	Creates concise, behavior-focused documentation for humans and AI. Use after feature implementation OR bug fixes to document HOW THE PRODUCT BEHAVES. NOT a changelog - documents current behavior, not change history.
allowed-tools	["Read","Grep","Glob","Write","Edit","mcp__ide__getDiagnostics"]

Creates concise, behavior-focused documentation that answers: **"How does this product/feature work?"**

This is NOT a changelog - it's an introduction to current product behavior.

Reference: See reference.md for templates, checklists, and examples.

**The 5-Year Reader Test**: Someone reading this in 5 years doesn't care that "we fixed a bug where X happened" - they just want to know how X works NOW.

Behavior Over History: Document what the product DOES, not what changed. When readers come back later, they need to understand current behavior, not archaeology.

Conciseness Over Completeness: A focused doc that gets read beats an exhaustive doc that gets skipped. Document what matters for understanding and usage.

Layered Documentation: Different scopes need different docs. System-level overviews, project guides, feature docs, and code comments each serve distinct purposes. See reference.md for layer definitions.

Documentation IS: Explaining WHY decisions were made, providing context for future changes, showing how pieces fit together, helping both humans and AI understand intent.

Documentation is NOT: A changelog of commits, implementation details without context, API reference without explanation.

<when_to_use> Features: After implementing significant new functionality (may span multiple commits) Bug Fixes: Update existing docs to reflect corrected behavior (don't add "bug fix" sections) Refactors: Only if external behavior or usage patterns changed

NOT for: Individual commits, internal refactors that don't change behavior, changelog entries </when_to_use>

<quick_start>

1. Understand feature scope - Review all commits related to the feature
2. Analyze architecture - Identify core types, data flow, design decisions
3. Generate docs/[feature-name].md - Use template from reference.md
4. Update package godoc - Use template from reference.md
5. Add type documentation - Use template from reference.md
6. Create testable examples - Example_* functions for complex types
7. Validate - Run through checklists in reference.md </quick_start>

<step_1_understand_scope>

• Review all commits related to the feature
• Identify all modified/new files
• Understand the problem being solved
• Map out integration points with existing system </step_1_understand_scope>

<step_2_analyze_architecture>

• Identify core domain types
• Map data/control flow
• Document design decisions (WHY choices were made)
• Note patterns used (vertical slice, self-validating types, etc.) </step_2_analyze_architecture>

<step_3_choose_documentation_layer> Before writing, decide where each piece of documentation belongs.

Decision Questions:

Question	If Yes →
Does this affect how multiple features interact?	System docs
Does this explain project setup, structure, or getting started?	Project docs
Does this explain how ONE feature works?	Feature docs
Does this explain ONE type/function's purpose?	Code docs

The Overlap Rule:

• Summarize up, detail down: Higher layers summarize, lower layers elaborate
• Good overlap: System doc mentions "auth uses JWT" → Feature doc explains JWT implementation
• Bad overlap: Same paragraph copy-pasted across multiple docs (will drift)

Cross-Reference, Don't Duplicate:

• System doc: "See docs/auth.md for authentication details"
• Feature doc: "See UserID godoc for validation rules"
• Code doc: "See docs/auth.md for architectural context"

See reference.md "Choosing Documentation Layer" for detailed decision tree. </step_3_choose_documentation_layer>

<step_4_generate_feature_doc> Create docs/[feature-name].md using the template in reference.md.

Cover:

• Problem & solution: What problem does this solve?
• Key players: Who are the main actors? (types, interfaces, services)
• Entry points: Where does execution start? (handlers, commands, events)
• Architecture: How does it work?
• Usage examples: How do I use it?
• Integration: How does it fit into the system?

For bug fixes: Don't create new docs. Update existing docs to reflect correct behavior. See reference.md "Bug Fix Documentation" guidelines. </step_4_generate_feature_doc>

<step_5_update_code_documentation> Using templates from reference.md:

1. Package godoc - Update to reflect feature's role
2. Type godoc - Explain purpose, design decisions, constraints
3. Function godoc - Only for non-obvious behavior
4. Testable examples - Example_* functions for complex types

Code docs should reference feature docs: See docs/[feature].md for detailed architecture </step_5_update_code_documentation>

<step_6_validate> Run through the checklists in reference.md:

• Feature documentation checklist
• Code comments checklist
• Quality gates (clarity, AI, maintenance, example tests)

Key questions:

• Can someone unfamiliar understand the feature?
• Can AI use this for bug fixes without reading all code?
• Are design decisions clearly explained?
• Are integration points documented? </step_6_validate>

<step_7_manage_size> If documentation grows too large (>500 lines), split into folder structure. See reference.md "Managing Documentation Size" for guidelines. </step_7_manage_size>

<output_format> After generating documentation:

DOCUMENTATION COMPLETE

Feature: [Feature Name]

Generated Artifacts:
- docs/[feature-name].md (created/updated)
- Package godoc updated in [package]/[file].go
- Type documentation for: [list types]
- Testable examples: [list Example_* functions]

Validation:
- [ ] Feature doc checklist passed
- [ ] Code comments checklist passed
- [ ] Quality gates passed

Next Steps:
1. Review docs/[feature-name].md for accuracy
2. Run `go test` to verify testable examples
3. Commit documentation

</output_format>

<success_criteria> Documentation is complete when ALL of the following are true:

Content:

• Feature doc created/updated at docs/[feature-name].md
• Package godoc updated to reflect feature's role
• Key types have godoc explaining purpose and design decisions
• Testable examples created for complex/core types

Quality:

• Describes current behavior (not history or changelogs)
• Someone unfamiliar can understand the feature
• AI can use this for bug fixes without reading all code
• Design decisions are clearly explained with rationale

Maintainability:

• Doc stays focused and scannable (split if >500 lines)
• No redundant information across layers
• Cross-references between doc layers are clear </success_criteria>