Suggest Skill

Proposes improvements and IMPLEMENTS SAFE ONES AUTOMATICALLY.

Autonomous Execution

DEFAULT BEHAVIOR: Implement safe improvements automatically.

Auto-Implement (no permission needed)

Implement automatically if ALL conditions are met:

• Low-to-medium effort (< 30 min work)
• Safe (doesn't change intended behavior)
• Clear benefit (readability, maintainability, performance)
• Reversible (easy to undo if wrong)

Examples of auto-implement:

• Extract repeated code into a function
• Rename unclear variables
• Add missing error handling
• Improve comment clarity
• Remove dead code
• Consolidate duplicate logic
• Add type annotations
• Fix inconsistent formatting

Pause for Human Decision

Present to user if ANY condition applies:

• High effort (> 30 min work)
• Behavior change (even if "better")
• Architectural (affects multiple components)
• Risky (could break things if wrong)
• Subjective (style preference, not clear improvement)

Examples requiring human:

• Rewrite module with different pattern
• Change API interface
• Add new dependency
• Restructure directory layout
• Change error handling strategy

Analysis Areas

Auto-Implement Categories

Code Quality (AUTO):

• Extract function from duplicated code
• Rename unclear variables/functions
• Remove unused imports/variables
• Add missing null checks
• Simplify complex conditionals

Maintainability (AUTO):

• Add missing docstrings to public functions
• Extract magic numbers to constants
• Group related code together
• Remove commented-out code

Performance (AUTO if safe):

• Cache repeated calculations
• Use more efficient data structures
• Remove unnecessary operations

Human Decision Categories

Architecture (PAUSE):

• Module restructuring
• Pattern changes
• New abstractions

Dependencies (PAUSE):

• Adding libraries
• Upgrading versions
• Removing dependencies

Behavior (PAUSE):

• Changing defaults
• Modifying error handling strategy
• Altering API contracts

Execution Flow

1. Analyze changes for improvement opportunities
2. Categorize each suggestion:
   - Safe + Low effort → AUTO-IMPLEMENT
   - Risky or High effort → PRESENT TO USER
3. Implement all auto-implement items
4. Run tests
5. If tests pass:
   - Report what was implemented
   - Present remaining suggestions to user
6. If tests fail:
   - Revert auto-implemented changes
   - Report the issue

Output Format

# Improvement Analysis

## Auto-Implemented
1. **[file:line]** Extracted `calculateTotal()` from duplicated code
   - Before: 15 lines repeated in 3 places
   - After: Single function, 3 call sites

2. **[file:line]** Renamed `x` to `userCount`
   - Improves readability

## Tests: PASS ✓

## Suggestions for Human Decision
1. **[HIGH IMPACT]** Refactor auth module to use middleware pattern
   - Effort: ~2 hours
   - Benefit: Cleaner separation, easier testing
   - Risk: Touches 8 files, could break auth flow
   - Recommendation: Yes, but schedule dedicated time

2. **[MEDIUM IMPACT]** Add Redis caching layer
   - Effort: ~4 hours
   - Benefit: 10x faster repeated queries
   - Risk: New dependency, operational complexity
   - Recommendation: Evaluate if performance is actually a problem

## Summary
- Analyzed: X opportunities
- Auto-implemented: Y
- Needs decision: Z

Integration with Process

After auto-implementing:

1. Tests are re-run automatically
2. If pass → continue to next phase
3. If fail → revert, report, pause for human

The process continues autonomously unless human decision is genuinely needed.

Anti-Patterns

DO NOT auto-implement:

• "Let's rewrite this in a better way" (subjective)
• "This could use a different pattern" (architectural)
• "We should add validation here" (behavior change)
• "Let's add logging everywhere" (scope creep)

DO auto-implement:

• "This variable name is unclear" → rename it
• "This code is duplicated" → extract it
• "This import is unused" → remove it
• "This null check is missing" → add it