AgentSkillsCN

iteration-controller

在设计评审周期中检测无限循环与陈旧迭代。通过对比各轮次的批评反馈，找出反复出现的问题。在管理迭代式评审流程时使用此功能，以避免在无法解决的问题上浪费时间与精力。

SKILL.md
--- frontmatter
name: iteration-controller
description: Detects infinite loops and stale iterations in design-review cycles. Compares Critic feedback across iterations to find repeated issues. Use when managing iterative review workflows to prevent wasted cycles on unresolvable issues.

Iteration Controller

When to use

  • •After receiving Critic verdict in iteration 2+
  • •When suspecting the design-review cycle is stuck
  • •Before submitting for another review iteration
  • •When iteration count approaches maximum (5)

How to use

Check for loops after a Critic review

bash
python3 .github/skills/iteration-controller/scripts/loop-detector.py check --folder generated_docs_[TIMESTAMP]

The script reads workflow_state.json history and compares Critic issues across iterations.

Output

json
{
  "iteration": 3,
  "max_iterations": 5,
  "loop_detected": true,
  "repeated_issues": ["Missing monitoring section", "Diagram uses C4 syntax"],
  "progress_score": 0.3,
  "recommendation": "ESCALATE - same 2 issues persist across 2+ iterations"
}

Interpret results

  • •loop_detected: false + progress_score > 0.5 → Continue normally
  • •loop_detected: false + progress_score < 0.5 → Progress slow, consider targeted fixes
  • •loop_detected: true → Alert user, show persistent issues, ask for guidance

How it works

  1. •Reads all Critic verdicts from workflow_state.json history
  2. •Extracts issue descriptions from each iteration
  3. •Compares issues using fuzzy string matching
  4. •Calculates progress_score = (resolved / total) across iterations
  5. •Flags loop if same issues appear in 2+ consecutive iterations