shinpr

commands-en

@shinpr/commands-en
shinpr
157
15 forks
Updated 1/18/2026
View on GitHub

Tool selection criteria (measurable decisions): Implement user skill change requests with maximum precision optimization

Installation

$skills install @shinpr/commands-en
Claude Code
Cursor
Copilot
Codex
Antigravity

Details

Path.claude/commands-en/refine-skill.md
Branchmain
Scoped Name@shinpr/commands-en

Usage

After installing, this skill will be available to your AI coding assistant.

Verify installation:

skills list

Skill Instructions


description: Implement user skill change requests with maximum precision optimization

Change request: $ARGUMENTS

Think deeply Extract the TRUE INTENT behind user's change request and implement with MAXIMUM PRECISION to eliminate ALL ambiguity:

9 Optimization Perspectives

  1. Context efficiency vs execution accuracy - Maximum accuracy with minimal description
  2. Deduplication - Consistency within and across skill files
  3. Proper responsibility aggregation - Group related content (minimize read operations)
  4. Clear decision criteria - Measurable standards
  5. Transform negatives to recommendations - Recommended format (background: including NG examples)
  6. Consistent notation - Unified expressions
  7. Explicit prerequisites - Make implicit assumptions visible
  8. Optimized description order - Most important first, exceptions last
  9. Clear scope boundaries - What's covered vs what's not

Execution Flow

1. Understand the Request

Question template when unspecified:

1. Which skill to modify?
   e.g.: typescript-rules / coding-standards / documentation-criteria

2. Select change type:
   a) Add new criteria (add new standards)
   b) Modify existing criteria (clarify ambiguous descriptions)
   c) Delete criteria (remove obsolete standards)

3. Specific changes:
   e.g.: "Add rule prohibiting 'any' type usage"
   e.g.: "Clarify error handling criteria"
   [User input]

2. Create Design Proposal

Target file identification and current state check:

# Tool selection criteria (measurable decisions)
if skill name is explicitly provided:
  Read: .claude/skills/{skill-name}/SKILL.md direct read
else if partial skill name known:
  Glob: .claude/skills/*{keyword}*/SKILL.md search
  Read: Check identified file's current state
else:
  Glob: .claude/skills/*/SKILL.md for full scan
  Confirm target skill selection with user

Design template:

【Current】
"Handle errors appropriately" (ambiguous: "appropriately" undefined)

【Understanding User Request】
"Want stricter error handling" → Set measurable criteria

【Proposal】
"Error handling implementation criteria:
1. try-catch required for:
   - External API calls (fetch, axios, etc.)
   - File I/O operations (fs.readFile, etc.)
   - Parsing operations (JSON.parse, parseInt, etc.)
2. Required error log items:
   - error.name (error type)
   - error.stack (location)
   - Timestamp (ISO 8601 format)
3. User notification criteria:
   - No technical details (NG: stack trace display)
   - Clear action items (recommended: "Please try again")"

Proceed with this design? (y/n)

3. Three-Pass Review Process

Pass 1: Add for Maximum Accuracy【Addition-Only Mode】

Declaration: "Pass 1: Addition mode execution. Deletions are prohibited."

  • Convert all ambiguous expressions → measurable criteria Example: "large" → "100+ lines" "5+ files"
  • Make all implicit prerequisites explicit Example: "In TypeScript environment" "Within async functions"
  • Define all exceptions and edge cases Example: "When null/undefined" "For empty arrays" Report: Count added items (minimum 5 items)

Pass 2: Critical Modification to Reduce Redundancy【Actual Modification Mode】

Declaration: "Pass 2: Critical modification mode execution. Actually delete and consolidate redundant parts."

Modification work:

  1. Critically review all Pass 1 additions
  2. Apply modifications using these criteria:
    • Duplicate concepts → Consolidate
    • Overly detailed explanations → Simplify
    • Overlap with other skills → Replace with references
  3. Record complete before/after diff (deletion reasons required)

Report format:

Modified locations: X items
Deleted/consolidated content:
- [Before]: "Detailed description"
  [After]: "Concise description"
  [Reason]: Redundancy elimination

Pass 3: Accuracy Assurance via Diff Evaluation【Restoration Decision Mode】

Declaration: "Pass 3: Diff evaluation mode execution. Review all Pass 2 modifications and restore if accuracy compromised."

Verification work:

  1. Evaluate each Pass 2 modification:
    • Does deletion create implementation ambiguity? → Restore
    • Are edge cases still covered? → Restore if missing
  2. Action mapping:
    • Deletions with accuracy risks → Restore
    • Valid deletions → Keep

Final confirmation (required answer): "Are necessary and sufficient conditions present for accurate implementation of user requirements?"

Report: Number of restored items and final reduction percentage

4. Get Approval

Present before/after comparison for user approval.

5. Implementation

  1. Apply changes with appropriate tool (after user approval)
  2. Final verification with git diff
  3. Suggest /sync-skills execution

Decision Criteria Checklist

  • Expressible in "if-then" format ("if X then Y")
  • Measurable by numbers/counts/states (eliminate subjective judgment)
  • Related content aggregated in single file (minimize read operations)
  • Relationships with other skills specified (dependencies/references/delegation)
  • NG examples included as background information
  • All prerequisites explicitly stated

Reduction Pattern Examples

PatternBeforeAfter
Emotional expressions"must always""execute when (background: build error if skipped)"
Time expressions"immediately""execute first after error detection"
Implicit prerequisites"implement error handling""TypeScript async function error handling"
Unclear order"consider: A, B, C""Priority: 1.A (required), 2.B (recommended), 3.C (optional)"
Redundant explanation"ensure type safety by defining types, checking types, and preventing type errors""type safety (define・check・prevent errors)"
Ambiguous scope"write tests""write unit tests (see typescript-testing skill for E2E tests)"

Output Example

=== Change Implementation Complete ===

【User Request】
"Strengthen TypeScript error handling criteria"

【Changes】
Target: .claude/skills/typescript-rules/SKILL.md
Section: ## Error Handling

Before:
"Handle errors appropriately"

After (3-pass review complete):
"Error handling implementation criteria:
1. try-catch block required for:
   - External API calls (fetch, axios, etc.)
   - File I/O operations (fs.readFile, fs.writeFile, etc.)
   - Exception-prone operations (JSON.parse, parseInt, etc.)
2. Required error log items:
   - error.name (error type)
   - error.stack (location)
   - new Date().toISOString() (timestamp)
3. User-facing messages:
   - No technical details (NG: stack trace display)
   - Clear action items (recommended: "Please try again")"

【Improvement Metrics】
- Decision clarity: 0% → 100% (all measurable)
- Ambiguous expressions: 1 → 0
- NG examples included as background

Run /sync-skills for metadata synchronization.

Execution Order

  1. Understand user request
  2. Analyze current state
  3. Design changes with 9 perspectives
  4. 3-pass review process
  5. User approval
  6. Apply changes
  7. Suggest sync-skills

Scope: Understanding user change requests and implementing with maximum accuracy

Error Handling

  • Skill not found: Display available skill list
  • Large change detected: Suggest phased implementation for 50%+ changes