Self-Improving Agent (tuituitu)

Self-improvement and learning skill for the agent. Activate when: - User gives feedback on agent's performance - Agent makes mistakes or errors - User asks a...

Security Scan

The skill's core behavior—reading and persistently updating the agent's MEMORY/AGENTS/SOUL files—is consistent with its stated self-improvement purpose, but vague instructions (and a referenced note about 'executing actual commands' and cron/heartbeat checks) give the agent broad discretion to modify persistent files and possibly perform system actions it didn't declare, which is concerning.

Purpose & Capability

The stated purpose (self-improvement, recording lessons and preferences) matches the instructions to read and update MEMORY.md and daily memory files. However, the reference doc explicitly advises 'Always execute actual commands, not just update files' and mentions heartbeat/cron jobs — capabilities that go beyond simple file edits and are not declared in requirements or install specs.

Instruction Scope

Runtime instructions give broad, immediate authority to update persistent files (MEMORY.md, memory/YYYY-MM-DD.md, AGENTS.md, SOUL.md) and to 'immediately update relevant files' without limits. They instruct the agent to read 'relevant memory files' (no path constraints) and actively detect patterns. The reference file suggests running system-level actions (cron/heartbeat, execute commands) which is not limited or qualified in SKILL.md—this vagueness grants the agent wide discretion over files and possible system operations.

Install Mechanism

Instruction-only skill with no install spec and no code files. Low surface area from an installation perspective — nothing is downloaded or written by an installer.

Credentials

The skill requests no environment variables or external credentials, which is appropriate. Still, it implies reading and writing workspace files (undisclosed paths), which can expose persistent state or secrets if those files exist in the agent's environment. The skill does not declare or restrict which files it may access.

Persistence & Privilege

The skill explicitly instructs the agent to modify persistent files that define memory and persona (MEMORY.md, AGENTS.md, SOUL.md). Those modifications can permanently change agent behavior. Although always:false reduces forced inclusion, the platform default allows autonomous invocation; combined with the ability to persist changes, this creates a risk that the agent could evolve in unexpected ways without user review. The reference's mention of cron/heartbeat implies potential ongoing background activity.

What to consider before installing

This skill will persistently modify the agent's memory and persona files, which can permanently change agent behavior. Before installing, decide whether you want the agent to auto-edit those files without approval. Recommended precautions: 1) Require user confirmation before any file changes; 2) Restrict writable paths (only a single approved memory file) and backup originals; 3) Disable autonomous invocation or set the skill to require explicit user invocation; 4) Review diffs of MEMORY.md/AGENTS.md/SOUL.md changes before applying or committing; 5) Ensure sensitive files are excluded from the agent workspace and that file permissions are locked down; 6) Ask the publisher for clarification: exactly which files will be modified, whether the skill will run commands/cron jobs, and whether writes can be scoped or reviewed. If you cannot get those assurances, treat the skill as higher-risk and avoid enabling persistent/autonomous behavior.

Self-Improving Agent Skill

Core Principle

Every interaction is a learning opportunity. The agent should actively improve itself based on feedback and experiences.

Learning Triggers

Activate this skill when:

1. User provides explicit feedback (positive or negative)
2. Agent makes an error or mistake
3. User says things like "learn from this", "remember", "don't do X again"
4. Agent notices a repeated mistake pattern
5. Task completes with lessons to extract

Self-Improvement Workflow

Step 1: Capture Learning

After any triggering event, immediately:

• Update MEMORY.md with new insights/lessons
• Update memory/YYYY-MM-DD.md with the event
• If mistake: add to "本猫犯过的错误/教训" section in MEMORY.md
• If preference: update "腿腿哥的新偏好/要求" section

Step 2: Reflect

Ask: "What can I learn from this?"

• What went wrong?
• What should I do differently?
• What files need updating?

Step 3: Act

Immediately update relevant files:

• MEMORY.md for long-term improvements
• Daily memory files for event-specific lessons
• AGENTS.md or SOUL.md if persona/tone changes needed

Step 4: Confirm

Tell the user what the agent learned and what changed.

File Update Guidelines

MEMORY.md Updates

Add to "本猫犯过的错误/教训" section:

- ❌ [Date]: [What went wrong]
  → 改进：[What to do differently]

Add to "腿腿哥的新偏好/要求" section:

- [New preference or requirement] (added [date])

Daily Memory Updates

Add to "做得不好/需要改进的地方":

- ❌ [What went wrong]: [Context]
  - 教训：[Learning]

Pattern Detection

Actively look for:

• Repeated mistakes (same error multiple times)
• User preferences that aren't recorded
• Missing knowledge that should be in memory
• Workflow inefficiencies

Quick Reference

When user says "learn from this":

1. Read relevant memory files
2. Identify the lesson
3. Update MEMORY.md immediately
4. Confirm update to user

When agent makes a mistake:

1. Acknowledge the mistake
2. Explain what went wrong
3. Update relevant files
4. Tell user what will change

---

This skill is part of the agent's self-improvement system.