AI Agent Hub
Back to skills
Agent Iteration Engine icon

Agent Iteration Engine

AI Agent Updated 2026.08.30

Paste the following prompt into your AI chat to install this skill:

Please install @user_ef64daef/diedai into your AI assistant according to https://skillhub.cn/install/skillhub.md.

About this skill

Problem

Many agents finish tasks but leave only transcripts behind. Errors, user corrections, and useful approaches are not retained, so similar failures recur. Dumping all history into memory files can create bloated rules, stale guidance, and conflicting instructions.

How It Works

  • Layered logging: records learning, error, and requirement entries, weighted by source; user correction outranks self-discovery, and one-off events rank lower.
  • Experience extraction: builds structured cards from tasks, conversation history, or MEMORY.md and TOOLS.md, with duplicate hints.
  • Promotion and conflict checks: qualified experiences can be written to SOUL.md, TOOLS.md, or AGENTS.md; contradictions require human review.
  • Lifecycle management: underused experiences decay and can be archived; search, growth reports, and export make the history inspectable.

Boundary

Best for iterative agent workflows that reuse rules and tool knowledge. It is not an automatic arbiter: promotion confirmations, thresholds, and summary quality still depend on operator judgment.

Use Cases

  • After an agent command fails or the user corrects it, capture the cause, context, and fix as searchable logs.
  • After a task, extract experience cards from `task.json`, chat history, or `MEMORY.md` to avoid repeating mistakes.
  • When a tool rule is repeatedly validated, evaluate whether it should be promoted into `TOOLS.md` or `AGENTS.md`.
  • As the experience base grows, run decay checks and growth reports to see which rules remain active or should be archived.

Best For

  • Engineers who want agents to retain tool pitfalls, fixes, and user corrections across tasks.
  • Agent developers maintaining rule files such as `SOUL.md`, `TOOLS.md`, or `AGENTS.md`.
  • Automation owners who want to turn task retrospectives into searchable, promotable, and archivable experience.
  • Team members reviewing agent growth reports and deciding which lessons should become formal rules.