API Documentation
Patterns for creating clear, complete API documentation that developers can use without guessing.
OpenAPI/Swagger spec completeness is deterministic — the linter script checks it exactly. Prose judgment (is this description clear? is this example realistic?) remains L2 work.
Purpose
- Provide consistent structure for API endpoint documentation
- Ensure all critical information (auth, errors, examples) is included
- Enable developers to make successful API calls on first try
Core Patterns
Pattern 1: Endpoint Specification
When to use: Documenting any REST API endpoint.
Implementation:
## `METHOD /path/:param`
Brief description of what this endpoint does.
### Authentication
Bearer token required in Authorization header.
### Parameters
| Name | Type | In | Required | Description |
|------|------|-----|----------|-------------|
| id | string | path | Yes | Resource identifier |
| limit | integer | query | No | Max results (default: 20) |
### Request Body
```json
{
"field": "value"
}
Response
{
"id": "abc123",
"created_at": "2024-01-01T00:00:00Z"
}
Errors
| Code | Meaning | Resolution |
|---|---|---|
| 400 | Invalid request | Check request body format |
| 401 | Unauthorized | Include valid Bearer token |
| 404 | Not found | Verify resource ID exists |
Example
curl -X GET "https://api.example.com/resource/abc123" \
-H "Authorization: Bearer TOKEN"
**Benefits:**
- Developers find information predictably
- All edge cases documented upfront
- Copy-paste examples reduce errors
### Pattern 2: Error Response Documentation
**When to use:** Documenting error responses for any endpoint.
**Implementation:**
```markdown
### Error Response Format
All errors return this structure:
```json
{
"error": {
"code": "RESOURCE_NOT_FOUND",
"message": "User with ID 'xyz' not found",
"details": {
"resource": "user",
"id": "xyz"
}
}
}
Common Error Codes
| Code | HTTP Status | Description | Action |
|---|---|---|---|
INVALID_REQUEST |
400 | Malformed request body | Validate JSON structure |
UNAUTHORIZED |
401 | Missing or invalid token | Refresh authentication |
FORBIDDEN |
403 | Insufficient permissions | Check user roles |
NOT_FOUND |
404 | Resource doesn't exist | Verify ID |
RATE_LIMITED |
429 | Too many requests | Wait and retry |
**Benefits:**
- Consistent error handling in client code
- Actionable resolution guidance
- Reduces support burden
---
## Anti-Patterns
### Anti-Pattern 1: Missing Authentication Details
| Aspect | Description |
|--------|-------------|
| **WHY** | Developers waste time debugging 401 errors; increases support tickets |
| **DETECTION** | Endpoint docs without "Authentication" section; no example with auth header |
| **FIX** | Always include auth method, header format, and token example |
**Bad Example:**
```markdown
## GET /api/users
Returns list of users.
### Response
```json
[{"id": "1", "name": "John"}]
**Good Example:**
```markdown
## GET /api/users
Returns list of users.
### Authentication
Requires Bearer token in Authorization header.
Authorization: Bearer
### Response
```json
[{"id": "1", "name": "John"}]
Example
curl -X GET "https://api.example.com/api/users" \
-H "Authorization: Bearer sk_live_abc123"
### Anti-Pattern 2: Undocumented Error States
| Aspect | Description |
|--------|-------------|
| **WHY** | Clients can't handle errors gracefully; leads to poor UX and debugging chaos |
| **DETECTION** | Docs only show success responses; no error codes or troubleshooting |
| **FIX** | Document every possible error with code, cause, and resolution |
**Bad Example:**
```markdown
### Response
```json
{"status": "success", "data": {...}}
**Good Example:**
```markdown
### Success Response (200)
```json
{"status": "success", "data": {...}}
Error Responses
| Status | Code | When | Resolution |
|---|---|---|---|
| 400 | INVALID_EMAIL |
Email format wrong | Use valid email |
| 409 | USER_EXISTS |
Email already registered | Use login endpoint |
| 422 | WEAK_PASSWORD |
Password too simple | 8+ chars, mixed case |
### Anti-Pattern 3: Example-Free Documentation
| Aspect | Description |
|--------|-------------|
| **WHY** | Forces developers to guess; increases time to first successful call |
| **DETECTION** | No curl/code examples; no sample request/response pairs |
| **FIX** | Include complete, copy-paste-ready examples for every endpoint |
**Bad Example:**
```markdown
## POST /api/users
Creates a new user with the provided data.
Good Example:
## POST /api/users
Creates a new user.
### Request
```bash
curl -X POST "https://api.example.com/api/users" \
-H "Authorization: Bearer TOKEN" \
-H "Content-Type: application/json" \
-d '{"email": "user@example.com", "name": "John Doe"}'
Response
{
"id": "usr_abc123",
"email": "user@example.com",
"name": "John Doe",
"created_at": "2024-01-15T10:30:00Z"
}
---
## Validation Checklist
### Pre-Documentation
- [ ] Endpoint actually exists and works
- [ ] Understand all parameters and their validation
- [ ] Tested all error cases
### Documentation
- [ ] Authentication section included
- [ ] All parameters documented with types
- [ ] Request body example provided (if applicable)
- [ ] Success response example included
- [ ] Error responses documented with resolution
- [ ] Complete curl example provided
### Post-Documentation
- [ ] Example actually works when copied
- [ ] All status codes verified against implementation
- [ ] Cross-linked to related endpoints
---
## Invocation — OpenAPI Coverage Linter (L3 Script)
When reviewing an OpenAPI 3.x or Swagger 2.0 spec, run the linter to get structured, ranked findings. Consume the script's **output only** — the script source never enters context.
**Input format:** JSON object — an OpenAPI or Swagger spec. YAML specs must be converted to JSON upstream (e.g., with `python3 -c "import sys, json, yaml; json.dump(yaml.safe_load(sys.stdin), sys.stdout)" < spec.yaml`).
**Rules checked:**
| Rule | Severity | What it checks |
|------|----------|---------------|
| R01 | MEDIUM | Operation missing `summary` [OAS3 §4.8.10.1] |
| R02 | LOW | Operation missing `description` [OAS3 §4.8.10.1] |
| R03 | MEDIUM | Path parameter missing description [OAS3 §4.8.12.1] |
| R04 | LOW | Query/header/cookie param missing description [OAS3 §4.8.12.1] |
| R05 | MEDIUM | Request body missing description [OAS3 §4.8.13.1] |
| R06 | LOW | Request body media type missing example [OAS3 §4.8.14.1] |
| R07 | MEDIUM | Response missing description [OAS3 §4.8.17.1] |
| R08 | LOW | Response schema present but no example [OAS3 §4.8.14.1] |
| R09 | HIGH | No 4xx response documented [RFC 7231 §6] |
| R10 | HIGH | Secured endpoint missing 401/403 response [OAS3 §4.8.21] |
| R12 | LOW | `operationId` not camelCase [OAS3 community convention] |
| R13 | LOW | `info` block missing `contact` or `license` [OAS3 §4.8.2] |
**Bash invocation (file argument):**
```bash
python .claude/skills/documentation/api-docs/scripts/api_coverage.py spec.json
Bash invocation (stdin):
cat openapi.json | python .claude/skills/documentation/api-docs/scripts/api_coverage.py -
Script output:
- A JSON block:
{ "spec_version": "oas3|swagger2", "findings": [...], "summary": { "total", "high", "medium", "low" } } - A markdown findings table sorted by severity (HIGH → MEDIUM → LOW).
What the agent does with the output:
- Read the HIGH-severity findings first — these represent auth/error documentation gaps that will cause client failures.
- Flag MEDIUM findings as required documentation debt.
- Present LOW findings as enrichment opportunities; leave the decision to fix them to the team.
- Reference the JSON summary counts in any executive-level documentation review.
Error handling: The script exits non-zero with an ERROR: message to stderr on bad JSON, non-object input, or unrecognised spec format. Fix the input and re-run.
Related Resources
- Related skills:
tutorial-patterns - OpenAPI Specification: https://swagger.io/specification/
Changelog
| Version | Date | Changes |
|---|---|---|
| 2.0.0 | 2026-05-20 | L3 script api_coverage.py added; Invocation section; allowed-tools updated |
| 1.0.0 | 2026-01-13 | Initial version with anti-patterns |