fresh-air

name	fresh-air
description	Recommends best-fit refactoring techniques for smell findings with confidence. Use when smell findings need scoped refactoring direction.
user-invocable	true
disable-model-invocation	false

Fresh Air

Core Contract

Recommend best-fit refactoring techniques for smell findings. Return one primary and optional fallback per smell for downstream planning. Do not author final plan files. Follow CLAUDE.md / AGENTS.md on conflict.

Required Inputs

1. Smell IDs.
2. Evidence location per smell.
3. Constraints and hard boundaries.

Workflow

1. Validate each smell entry (smell_id, evidence_location, constraints).
2. Load reference mappings (smells, technique-map, techniques datasets).
3. Rank techniques by evidence fit, constraints, risk, and reversibility.
4. Return per-smell recommendation block:
   • primary technique + rationale,
   • optional fallback + trigger,
   • concrete improvement suggestion,
   • confidence.

Stop and ask when evidence is missing or no mapped technique exists.

Safety Rules

• Never drop smells silently; if capacity is exceeded, process in explicit chunks and report remaining count.
• Never recommend techniques without mapping evidence unless the user approves a fallback strategy.
• Never skip loading technique detail files for top candidates when they exist.
• If a required technique detail file is missing, state reduced confidence and ask whether to proceed with metadata-only ranking.
• Never hide low confidence; call it out and ask a targeted clarifying question.
• Never produce architecture-wide recommendations when a local technique is sufficient.

Output Style

Report shared constraints, compact per-smell recommendations, confidence per smell, and one optional clarification question when needed.