Self-Improving Agent

pskoettAgentsOpenClaw

Capture errors and corrections, maintain reusable learnings and review them before future work.

Illustrative Self-Improving Agent workflow: Error + context + correction; example output includes LEARNINGS.md, ERRORS.md, FEATURE_REQUESTS.md.

Install Self-Improving Agent

Use OpenClaw’s built-in skill installer to install the publisher’s skill into the active workspace.

Install into the active OpenClaw workspace

openclaw skills install @pskoett/self-improving-agent

Copying the command does not install the skill. Review the publisher’s source and workspace requirements before running it.

What this skill helps with

pskoett’s Self-Improvement skill for OpenClaw maintains structured records of learnings, errors and feature requests. It covers recording corrections, reviewing saved guidance and promoting reusable knowledge when its scope is understood.

When to use it

Use it after a command failure, a corrected assumption, a missing capability or a repeated workflow problem. Relevant saved guidance should be reviewed and revalidated before it is reused.

What to bring

Provide the error or correction, enough task context to explain it and the intended workspace. Prefer short, sanitised evidence, and identify the scope where the learning applies.

What to expect

Expect learning records, error entries or requested-capability notes, with reusable guidance maintained separately. This is a file-based workflow; the page does not change an agent’s model weights or automatically improve its performance.

License information

The verified source folder does not publish a license file. Check the publisher’s current terms before reusing or redistributing the skill; this page does not assign it an open-source license.

How the skill works

Capture the relevant event

Place corrections, failures and missing-capability requests in the corresponding learning files, preserving existing records.

Review scope and evidence

Search for related entries, distinguish a new occurrence from a duplicate and revalidate claims when the underlying task or dependency changes.

Promote useful guidance carefully

Move broadly applicable guidance into the appropriate workspace instructions when justified. Optional hooks and cross-session sharing need their own explicit setup.