sc-patterns

sc-patterns: Codebase Pattern Detection and Convention Guidance

Purpose: Extract high-confidence coding conventions from a codebase and make them available as planning constraints for AI agents.

Mode Detection

Parse $ARGUMENTS to determine the mode:

Input	Mode	Action
Empty or analyze	Analyze	Full pattern scan, cache results
brief <task description>	Brief	Task-scoped pattern constraints from cache
init	Init	Analyze + generate starter policy.yaml
refresh	Refresh	Force full re-analysis, ignore cache

Mode: Analyze (default)

Run the sc-pattern-analyzer agent to scan the codebase.

Steps:

1. Check if .patterns/evidence.json exists and is fresh:
   • Read .patterns/.cache-meta.json for the last-analyzed commit SHA
   • Compare to current HEAD SHA via git rev-parse HEAD
   • Check if working tree has changes in analyzed paths via git status --porcelain
   • If cache is fresh and working tree is clean in analyzed scopes: report cached results instead of re-scanning
2. If stale or missing: spawn sc-pattern-analyzer agent with the full analysis task
3. After analysis completes, display a human-readable summary:
   • Count of patterns by confidence band (enforceable / probable / weak / conflicted)
   • Top enforceable patterns with golden file references
   • Any "no pattern detected" scopes
   • Policy overrides applied (if .patterns/policy.yaml exists)

Output example:

## Pattern Analysis Complete

Analyzed 342 files (89 excluded) | Primary: Ruby/Rails

### Enforceable (8 patterns)
- Service objects use `def call` pattern (92% in app/services/)
- Controllers use before_action guards (88% in app/controllers/)
- Models use scopes over class methods (78% in app/models/)
  ...

### Probable (4 patterns)
- Repository pattern for DB access (65% in app/repositories/)
  ...

### Conflicted (2 areas)
- Error handling in services: 40% Result objects, 35% exceptions, 25% booleans
- Component styling: 55% CSS modules, 45% styled-components

### No Strong Pattern
- Serializer conventions in app/serializers/

Cache written to .patterns/evidence.json
Brief written to .patterns/brief.json

Mode: Brief

Generate task-scoped pattern constraints from cached analysis.

Steps:

1. Read .patterns/evidence.json (run analyze first if missing/stale)
2. Parse the task description to identify relevant scopes:
   • Extract mentioned directories, file types, or entity types
   • Map entity types to directory scopes (e.g., "service" -> app/services/)
   • Include patterns from the framework/language scope
3. Filter patterns to only those relevant to the task's scope
4. Include only enforceable and probable confidence patterns
5. Read .patterns/policy.yaml for any overrides affecting the task scope
6. Output a focused pattern brief

Output example for /sc-patterns brief add a new user registration service:

## Pattern Brief: User Registration Service

Relevant scope: app/services/, app/models/, spec/services/

### Follow These Patterns

1. **Service objects use `def call`** (enforceable, 92%)
   See: app/services/create_user_service.rb:8-22

2. **Services accept a params hash** (enforceable, 88%)
   See: app/services/update_profile_service.rb:5-12

3. **Tests mirror source structure** (enforceable, 95%)
   Place test at: spec/services/register_user_service_spec.rb

### Avoid

- Do NOT use multiple public methods (legacy pattern in legacy_export_service.rb)
- Do NOT raise exceptions for business logic failures (per policy.yaml — use Result objects)

### No Consensus (your judgment)

- Error return format: 40% Result objects, 35% exceptions. Policy says use Result objects.

Mode: Init

Run analysis and generate a starter .patterns/policy.yaml for human review.

Steps:

1. Run the full analyze flow
2. Generate .patterns/policy.yaml pre-populated with:
   • Detected enforceable patterns as enforce: entries (for human confirmation)
   • Detected conflicted areas as comments asking for human decision
   • Standard exclude_paths for the detected framework
   • Placeholder boundaries section
3. Tell the user to review and commit the policy file

Output:

## Pattern Detection Initialized

Analysis complete. Generated:
- .patterns/evidence.json (full analysis, gitignored)
- .patterns/brief.json (compact brief, gitignored)
- .patterns/policy.yaml (human overrides — review and commit this)

Next steps:
1. Review .patterns/policy.yaml — confirm, edit, or remove entries
2. Commit policy.yaml to your repo
3. Run /sc-patterns analyze after significant codebase changes
4. Use /sc-patterns brief <task> before planning new features

Mode: Refresh

Force a full re-analysis regardless of cache state. Same as analyze but skips the freshness check.

Cache Management

The .patterns/ directory at project root stores analysis artifacts:

.patterns/
  evidence.json       # Full analysis — all patterns, all confidence levels
  brief.json          # Compact — enforceable + probable only, for prompt injection
  policy.yaml         # Human overrides — committed to repo
  .cache-meta.json    # Fingerprint: HEAD SHA, analyzed paths, detector version

Gitignore guidance: evidence.json, brief.json, and .cache-meta.json should be gitignored. policy.yaml should be committed. On first init, check if .gitignore includes .patterns/ entries and suggest additions if missing.

Integration Notes

This command writes files that sc-plan and sc-work can read. The integration is file-based, not import-based:

• sc-plan/sc-work check if .patterns/brief.json exists
• If it exists and is fresh, they read it for pattern context
• If missing, planning proceeds normally without pattern constraints
• No plugin dependency — just a file on disk

---

User Request: $ARGUMENTS