Quriosity-agent

videocut-self-evolve

Self-evolving skills. Record user feedback, update methodology and rules. Triggers: update rules, record feedback, improve skill, 更新规则, 记录反馈, 改进skill

Quriosity-agent 26 4 Updated 3mo ago

Resources

1
GitHub

Install

npx skillscat add quriosity-agent/qcut/videocut-self-evolve

Install via the SkillsCat registry.

SKILL.md

Self-Evolve

Let the Agent learn from mistakes and continuously improve

Quick Start

User: Record the issue we just had
User: Update the speech error detection rules
User: Remember this lesson
User: 记录一下刚才的问题
User: 更新口误识别的规则

Update Locations

Content Type Target File Example
User profile CLAUDE.md Preferences, habits
Methodology + feedback */tips/*.md Rules, lessons

Flow

User triggers ("that just failed", "record this")
    ↓
[Auto] Trace back context, find the problem
    ↓
[Auto] Read target file fully, understand structure
    ↓
[Auto] Integrate into the appropriate section (not just append!)
    ↓
[Auto] Feedback log only records events, doesn't repeat rules
    ↓
Report update results

Key: Don't ask "what's the problem" — analyze from context directly!

Update Principles

Bad: Append to the end

## Feedback Log
### 2026-01-14
- Lesson: Must generate deletion task list at end of review
- Lesson: Confirm errors and silence separately

Only adding to feedback log = rules scattered at the bottom, will repeat mistakes.

Good: Integrate into body

  1. Read full file, understand section structure
  2. Find the right section, integrate the rule
  3. Feedback log only records events: - Review marked silence, but missed during cut
## IV. Review Format
(Add deletion task list template)

## V. Confirmation & Execution Flow  ← Add this section if missing
(Add separate confirmation for errors and silence)

## Feedback Log
### 2026-01-14
- Review marked silence, but missed during cut (only cut errors)

Trigger Conditions

  • User corrects AI mistakes
  • User says "remember this", "watch out for this in the future"
  • A new general pattern is discovered

Anti-patterns

2026-01-13

Bad:
User: That just failed, update the skills
AI: What problem did you find?  ← Should NOT ask!

Good:
AI: [Auto trace back context, find failure point]
AI: [Execute update]

2026-01-14

Bad:
AI: Updated. Added 3 lessons to feedback log  ← Only appended!

Good:
AI: [Read full file, understand structure]
AI: [Integrate into appropriate sections]
AI: [Feedback log only records events]
AI: Updated: Added Chapter V "Confirmation Flow", updated Chapter IV template

Principle: Rules must be integrated into the body; feedback log is just an event journal.