Improve Skill

Input: $ARGUMENTS

Core Principles

1. Read before diagnosing. Never improve a skill you haven’t read in full. The most common failure is proposing generic improvements to a skill whose actual content you don’t understand. Read every line before generating a single critique.

2. The quality standard is objective, not aesthetic. A skill is good or bad relative to specific structural requirements: core principles, failure modes table, depth scaling table, pre-completion checklist, integration section. Missing elements are defects, not style choices.

3. Content quality is domain-specific. Generic principles (“be thorough”, “consider alternatives”) indicate a skill that doesn’t understand its own domain. Good principles name specific failure modes and specific interventions that only apply to THIS skill’s problem space.

4. Improvement means closing gaps, not rewriting from scratch. If a skill has strong content but missing structure, add the structure. If it has good structure but generic content, sharpen the content. Identify what’s weak and fix THAT — don’t destroy what works.

5. The exemplar skills define the floor. Skills like /foht (300 lines), /sbfow (279 lines), /iterate (373 lines), /araw (410 lines) represent the quality standard. Any improved skill must be comparable in depth, specificity, and structural completeness.

Phase 1: Read and Baseline

Read the target skill completely.

[I1] SKILL: /[name]
[I2] CURRENT_LINES: [line count]
[I3] CURRENT_STRUCTURE: [list which standard elements exist and which are missing]

Structural Audit

Check for each required element:

Phase 2: Content Diagnosis

Evaluate content quality beyond structure.

[I4] PRINCIPLES_QUALITY: [are principles domain-specific or generic platitudes?]
     GENERIC_PRINCIPLES: [list any that could apply to ANY skill — these need replacement]
     MISSING_INSIGHTS: [what domain-specific knowledge is absent?]

[I5] PHASE_QUALITY: [do phases have concrete steps or vague instructions?]
     VAGUE_PHASES: [list any that say "analyze" or "consider" without specifying HOW]
     MISSING_PHASES: [what steps are implied but not explicit?]

[I6] FAILURE_MODES_QUALITY: [are failure modes specific to this skill's domain?]
     GENERIC_FAILURES: [list any that apply to everything — "didn't think hard enough"]
     MISSING_FAILURES: [what domain-specific failures are unaddressed?]

[I7] OUTPUT_QUALITY: [is the output format specific and useful?]
     MISSING_OUTPUT: [what should the output include that it doesn't?]

Content Quality Tests

Phase 3: Improvement Plan

Generate specific fixes, not vague suggestions.

[I-N] FIX: [specific change]
     TYPE: [structural | content | integration | clarity]
     PRIORITY: [critical | important | nice-to-have]
     CURRENT: [what exists now — quote or describe]
     PROPOSED: [what should replace it — be specific]
     RATIONALE: [why this improves the skill]

Fix Priority Rules

• Critical: Missing required structural element (no failure modes table, no depth scaling)
• Critical: Generic principles that could apply to any skill
• Important: Vague phases that don’t specify concrete steps
• Important: Missing domain-specific failure modes
• Nice-to-have: Formatting improvements, better examples, additional integration links

Phase 4: Apply Fixes

Write the improved skill. Apply all critical and important fixes. Apply nice-to-haves if they don’t bloat.

Application Rules

• Preserve existing content that’s already good — don’t rewrite what works
• Add missing structural elements in their standard positions
• Replace generic principles with domain-specific ones
• Ensure the improved skill is 150-300 lines (the quality range)
• Verify every phase has concrete, followable steps

Phase 5: Verify

Re-run the structural audit and content diagnosis on the improved version.

[I-N] VERIFICATION:
     STRUCTURAL_GAPS_REMAINING: [should be zero]
     GENERIC_CONTENT_REMAINING: [should be zero]
     LINE_COUNT: [should be 150-300]
     ALL_TESTS_PASS: [disagreement, domain, specificity, failure, actionability]

Failure Modes

Depth Scaling

Default: 2x. These are floors.

Pre-Completion Checklist

• Target skill read in full before any diagnosis
• Structural audit completed with all elements checked
• Content diagnosis identifies specific (not generic) issues
• Every fix has current/proposed/rationale
• Critical fixes all applied
• Improved skill has all required structural elements
• Improved skill principles pass the domain test
• Line count is in the 150-300 range
• Verification confirms zero structural gaps

Integration

• Use from: skill maintenance, quality audits, skill expansion tasks
• Routes to: /w (if skill needs actual prose improvement), /cs (if skill needs structural rework)
• Complementary: /impss (improves multiple skills), /imprt (auto-identifies what to improve)
• Differs from /cs: cs creates new skills; imps improves existing ones
• Differs from /impss: impss handles batches; imps handles one skill deeply
• Differs from /fmtsb: fmtsb formalizes drafts into production; imps improves already-production skills