AgentSkillsCN

AI Feedback Reporter

关键提示——在任何文件写入操作之前,请先对已加载的技能进行违规审查。可通过反馈命令自动上报。

SKILL.md
--- frontmatter
name: AI Feedback Reporter
description: CRITICAL - Before ANY file write, audit loaded skills for violations. Auto-report via feedback command.
metadata:
  labels: [meta, feedback, quality, monitoring]
  triggers:
    files: ['**/*']
    keywords: [write, edit, create, generate, skill, violation]

Feedback Reporter

Priority: P0 - Auto-detect skill violations before file writes

🚨 Checkpoint: Before File Writes

Quick check before write_to_file, replace_file_content, multi_replace_file_content:

  1. Check - Any skills loaded for this file extension?
    • NO → ✅ Skip, proceed safely
    • YES → Continue to step 2
  2. Audit - Does planned code violate loaded skill rules?
    • NO → ✅ Proceed
    • YES → Run feedback command, then fix

Detection Flow

code
Before file write?
├─ Check file extension → Identify loaded skills
├─ Review skill anti-patterns/rules
├─ Code matches anti-pattern?
│  ├─ YES → 🚨 VIOLATION → feedback command
│  └─ NO → ✅ Proceed
└─ No skills loaded → ✅ Proceed

Examples (Quick Reference)

Flutter: color: Colors.blue → Rule: No hardcoded colors → Report + use theme

React: class MyComponent extends... → Rule: Use functions → Report + convert

SKILL.md: 85 lines → Rule: ≤70 max → Report + extract to references/

More examples →

Command (Copy-Paste)

bash
npx agent-skills-standard feedback \
  --skill="[skill-id]" \
  --issue="[violation]" \
  --skill-instruction="[exact rule]" \
  --actual-action="[what I did]"

Pre-Completion Check

Before notify_user or task completion:

Did I write code? YES → Did I audit skills? NO → Audit now

Anti-Patterns

  • No "I'll check later": Check before writing, not after
  • No "minor change skip": Every write needs check
  • No "user waiting skip": 10-second check > pattern violation