44 lines
1.2 KiB
Markdown
44 lines
1.2 KiB
Markdown
# Phase 14: Skill Sharpener
|
|
|
|
**Purpose:** Meta-review that evaluates execution results and generates improvements.
|
|
**TS Module:** `src/tier3-compounding/skill-sharpener.ts`
|
|
|
|
## What It Does
|
|
|
|
After each skill execution, the sharpener:
|
|
1. Checks execution success — did all steps pass?
|
|
2. Evaluates quality score — is it above 0.5, 0.8?
|
|
3. Analyzes duration trends — is this run slower than average?
|
|
4. Reviews improvement suggestions — what did the execution self-report?
|
|
5. Runs full evolution analysis — what structural changes are needed?
|
|
|
|
## Review Score
|
|
|
|
```
|
|
Score = 50 (base) + qualityScore * 30 - failurePenalty + changeBonus
|
|
```
|
|
|
|
- Score < 30: no actionable improvements
|
|
- Score 30-70: minor improvements available
|
|
- Score > 70: significant improvements recommended
|
|
|
|
## Quality Gate
|
|
|
|
Improvements are only applied if:
|
|
1. Review score >= 30
|
|
2. At least one suggested change exists
|
|
3. The change is not a duplicate of an existing rule
|
|
|
|
## Usage
|
|
|
|
```bash
|
|
fable-agent skills sharpen
|
|
```
|
|
|
|
## Key Rules
|
|
|
|
- Run sharpen after every 3 skill executions
|
|
- The sharpener never modifies skills directly — it generates diffs
|
|
- Applied changes bump the skill version automatically
|
|
- Review history is persisted in `~/.fable-agent/reviews/`
|