---

name: skillfit-optimizer description: >- Build and optimize a minimal working skill stack for a user goal: recommend profiles, install the best-fit stack, run deterministic smoke checks, remove overlap, score stack quality, persist before/after evidence, and produce a fix-first rollout plan. Use when users ask what skills to install, how to avoid duplicate/conflicting skills, how to reduce setup friction, or how to improve skill ROI/time-to-value.

SkillFit Optimizer

Objective

Reduce time-to-value from skill discovery by turning goals into a tested, minimal, high-ROI skill stack with measurable before/after evidence.

Typical Trigger Phrases

• "what skills should I install for X"
• "optimize my current skill stack"
• "remove duplicate/conflicting skills"
• "which skill setup gives best ROI"

Workflow

1. Analyze goal and context

• Parse user objective, constraints, and required outputs.
• Identify must-have capabilities and optional enhancements.

2. Propose 3 stack profiles

• Minimal (lowest complexity)
• Balanced (default)
• Maximum (capability-rich)

3. Install selected profile

• Prefer fewer, higher-signal skills.
• Avoid overlapping tools unless explicit fallback needed.

4. Smoke-check stack (deterministic)

• Run scripts/stack_check.py --bins <list> --history-path .skillfit/history.json --report-json .skillfit/latest-check.json.
• Validate one realistic happy-path task.
• Enforce gate: if availability_score < 80, do not claim success; return explicit fix commands first.

5. Prune overlap

• Remove redundant skills and conflicting patterns.
• Keep one primary path per capability.

6. Score stack quality (0-100)

• Coverage (0-30)
• Reliability (0-30)
• Setup friction (0-20, inverse)
• Overlap discipline (0-20, inverse penalty)

7. Persist + recurrence loop

• Persist each run in .skillfit/history.json.
• Track recurring issues (same missing bins / same overlap class).
• If a blocker recurs 3+ times in 30 days, elevate as high-priority remediation.

8. Promotion rule

• If a repeated pattern becomes generally useful, promote concise rule(s) to:
  • AGENTS.md for workflow safeguards
  • TOOLS.md for local tool gotchas
  • SOUL.md for behavioral defaults (when applicable)

Required Output Structure

1. Goal Fit Summary
2. Recommended Profile (Minimal / Balanced / Maximum)
3. Installed Stack
4. Smoke Check Results (include deterministic checker output)
5. Pruned/Removed Items
6. Stack Score + Rationale
7. Exact Fix Commands (copy/paste)
8. Next 3 Improvements
9. Before/After Delta (availability, missing bins, overlap)

Quality Rules

• Prefer execution certainty over skill quantity.
• Do not keep duplicate skills with same core function unless user requests redundancy.
• Flag unresolved setup blockers explicitly.
• If smoke check fails, return fix-plan before claiming success.
• Always provide deterministic evidence (checker output + missing bins list + delta vs previous run).

Reference

• Read references/profile-templates.md for profile patterns and scoring details.
• Read references/ops-report-template.md for report format and gate language.