Self-Improve Skill Skill

AI-consumed reference. Optimized for Claude to read during execution. Human-readable explanation: see docs/architecture/HIERARCHICAL_PLANNING.md or docs/getting-started/ depending on topic.

Self-Improve Skill

Apply learned improvements: update rules, adjust agent routing, modify workflow configs, generate knowledge entries.

Usage

/af learn apply                    # Review and apply pending
/af learn apply --auto             # Auto-apply high-confidence (>=0.8)
/af learn apply --preview          # Preview without applying
/af learn apply --id <pattern_id>  # Apply specific pattern

Improvement Types

types[4]{type,target,example}:
  Rule updates,rules/*.md,Increase coverage threshold 80→85
  Agent routing,agent-detector config,Default react-expert for .tsx
  Workflow adjustments,workflow config,Increase Phase 2 timeout
  Knowledge base,knowledge entries,TDD reduces bugs by 40% for APIs

Safety Guards

Approval required unless: --auto AND confidence >= 0.8 AND frequency >= 5.

Rollback: Every change creates backup + log. /af learn rollback <id> or --all.

Validation: Syntax check, conflict detection, impact assessment before applying.

Apply Process

Fetch: Query v_improvement_suggestions WHERE applied = FALSE
Generate: Determine target files, create modifications, calculate impact
Review: Present diff with confidence, frequency, evidence. User chooses: Apply / Skip / Modify
Apply: Create backup, apply modification, mark applied in Supabase, log change

Rollback

/af learn rollback <change_id>     # Specific change
/af learn rollback --list          # List recent changes
/af learn rollback --all           # All changes from today

Configuration

learning:
  self_improve:
    enabled: true
    auto_apply_threshold: 0.8
    min_frequency: 5
    backup_dir: backups/
    max_auto_per_day: 10

Agent Skills: Self-Improve Skill

Install this agent skill to your local