- Home
- Skills
- Prompt & Skill Engineering
- Self-Improving Agent
Works with the AI Agents you already use
Self-Improving Agent
Your agent learns from every correction and failed command: verified one-line rules in LEARNINGS.md, applied at session start, pruned and promoted to CLAUDE.md/AGENTS.md. Never repeats the same mistake twice.
Free
Self-Improving Agent
Example session with this skill installed
I keep getting timeouts on the database tests because they run in parallel. From now on, use --runInBand for pnpm test. Learn from this.
- Read your context and instructions
- Compiled the self-improving agent
I have verified the fix and updated your project learnings.
Learning Capture
Learned
- [test] Run 'pnpm test --runInBand' for db tests; parallel runs deadlock on the shared Postgres. (why: user correction after timeouts; seen: 2026-11-10; hits: 1)
Next steps
- I will automatically apply the
--runInBandflag for all future test executions in this project. - If you notice this rule becoming obsolete after a database upgrade, tell me to "prune learnings."
- I can promote this to your
CLAUDE.mdif it consistently prevents failures across 3 or more sessions.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop repeating the same mistakes across development sessions. This skill enables your agent to capture hard-won fixes, user preferences, and environment quirks into a persistent LEARNINGS.md file. It turns every failed command or manual correction into a durable, machine-readable rule that is automatically reviewed at the start of new tasks.
What it does
- Automatic capture identifies lessons from failed tests, build errors, and direct user feedback.
- Rule verification ensures only working solutions are stored by re-running commands before saving.
- Contextual review reads and applies relevant project-specific rules at the start of every session.
- Learning maintenance prunes stale entries and promotes frequent rules to main configuration files.
- Conflict resolution detects and supersedes contradictory rules based on new evidence.
How it works
- Trigger the skill after a fix or when you tell the agent "remember this."
- Verify the lesson by testing the specific command or configuration change.
- Record the finding as a structured, one-line rule in
LEARNINGS.mdwith category and hits. - Apply the knowledge automatically in future sessions by checking the file for relevant tags.
Frameworks & tools
This skill is framework-agnostic and works with any codebase. It utilizes LEARNINGS.md for project scope and ~/.agent/LEARNINGS.md for global preferences. It integrates with CLAUDE.md or AGENTS.md for high-frequency rule promotion.
Why this beats prompting it yourself
Standard prompts forget the context of yesterday's debugging session. This skill creates a verified knowledge base that grows with your repo, ensuring the agent doesn't need to be told twice about your specific CI quirks or preferred pnpm flags.
Use cases
- Documenting brittle test suites that require specific flags to avoid deadlocks.
- Storing project-specific architectural preferences that differ from LLM defaults.
- Remembering undocumented environment variables or setup steps for legacy apps.
- Tracking version-specific quirks for internal libraries or obscure APIs.
Known limitations
Does not store secrets or temporary data like branch names. Designed for short, imperative rules rather than long-form documentation. Effectiveness depends on the agent's ability to read the file.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Excludes secrets, tokens, and temporary data (ports/branches). File size is capped at 150 lines for performance. Requires manual confirmation for user preference updates.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
This skill has been used 7 times.
Trust & safety
Security scan
Security scan passed 7 days ago
- Free with an Agensi account