self-improve
Transforms simple user requests into enriched descriptions with traceable findings. Use when interpreting vague user intent, preparing context for agent generation, or synthesizing actionable insights from available project context.
When & Why to Use This Skill
The self-improve Claude skill is a powerful utility designed to bridge the gap between ambiguous user prompts and actionable agent execution. By synthesizing available project context and extracting traceable findings, it transforms simple requests into enriched, high-context descriptions. This skill optimizes agentic workflows by ensuring that intent is clearly interpreted and complexity is assessed before task execution, significantly improving the accuracy and relevance of AI-generated outputs in complex environments.
Use Cases
- Interpreting Vague Intent: Automatically expanding brief or ambiguous user requests into detailed project requirements by scanning available documentation and identifying implicit needs.
- Agent Context Preparation: Gathering and linking relevant source materials to provide a comprehensive, traceable knowledge base for downstream agent tasks like code generation or technical writing.
- Complexity Assessment: Evaluating project scope, identifying conflicting findings, and highlighting technical trade-offs to determine if a task requires deeper manual analysis or can be automated.
- Actionable Insight Synthesis: Converting raw project data and scattered context into a structured format with 'Discovery → Source' mapping, ensuring every recommendation is verifiable and actionable.
| name | self-improve |
|---|---|
| description | Transforms simple user requests into enriched descriptions with traceable findings. Use when interpreting vague user intent, preparing context for agent generation, or synthesizing actionable insights from available project context. |
| allowed-tools | Read |
Self-Improve
Transform user intent into actionable understanding.
Core Function
User Intent + Available Context → Enriched Description
Behavior
- Specific intent → focused interpretation, targeted findings
- Vague intent → broader scan, note what clarification needed
Process
1. Interpret Intent
Extract from user request:
- Action: What to do
- Subject: What to affect
- Implicit: What's assumed but not said
2. Synthesize Findings
Connect intent with relevant context. Each finding:
[Discovery] → [source]
Requirements: traceable, relevant, actionable.
3. Assess Complexity
| Factor | Question |
|---|---|
| Scope | How much affected? |
| Conflicts | Findings contradict? |
| Trade-offs | Significant choices? |
Any factor high → flag for deeper analysis.
Output
## Intent
[Interpreted user intent]
## Findings
1. [Discovery] → [source]
2. [Discovery] → [source]
3. [Discovery] → [source]
## Complexity
[Simple | Needs analysis: scope/conflicts/trade-offs]