agentsclimarketplace
Sign in

Ai memory assistant

Skill kelongyan/ai-memory-plugin/skills/ai-memory-assistant

This skill should be used when the user asks to "查看我的习惯", "回顾最近失败的命令", "总结 ai-memory 记录", "为什么这个命令总失败", "分析 ~/.ai-memory", or wants Claude Code to interpret remembered command history, lessons, and recurring failures stored under ~/.ai-memory.From its SKILL.md

Install
npx -y skills add kelongyan/ai-memory-plugin --skill ai-memory-assistant

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.1 KB, as published. Nobody here has run it

AI Memory Assistant

Purpose

Read the user's shared memory files under ~/.ai-memory/ and explain the patterns in natural language.

Focus on:

  • recurring Bash command failures
  • repeated error signatures
  • reusable lessons already extracted by the plugin
  • stable command habits reflected in preferences or stats
  • differences between global allow candidates and project-specific allow candidates

Workflow

  1. Read the relevant files in ~/.ai-memory/:
    • lessons.json
    • stats.json
    • events.jsonl when recent raw events matter
    • preferences.json when user asks about habits or preferences
  2. Prefer summarized files (lessons.json, stats.json) before reading raw events.
  3. When the user asks about habits, inspect preferences.json and distinguish:
    • global allow candidates
    • project allow candidates (scope=project, with cwd)
    • whether a suggestion is broadly stable or cwd-specific
  4. Explain findings in concise Chinese, but keep commands, paths, and raw error strings in original form.
  5. Distinguish between:
    • confirmed repeated patterns
    • one-off failures
    • suggestions that still need user confirmation
  6. If the user asks how to improve the memory rules, propose changes conceptually unless they explicitly ask for implementation.

Output guidance

When summarizing memory, structure the response as:

  • Observed patterns
  • Likely causes
  • Reusable advice
  • What is still uncertain

Guardrails

  • Do not claim a pattern is stable unless the data shows repeated events.
  • Do not suggest automatically changing Claude Code permissions unless the user explicitly asks.
  • Treat events.jsonl as raw evidence and lessons.json as derived guidance.

Keep looking

Skills are one crate of 627,559. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.