> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cogniagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Skills that improve themselves
> Your AI coworker proposes new skills and tidies up old ones based on what it learned. Nothing changes until you approve it.
Your coworker notices what it learned and proposes better skills. You decide what actually changes.
## What it does
After enough work, patterns show up. The same correction three tasks running. A method that worked well and is worth keeping. Two skills that have drifted into saying almost the same thing.
Your coworker spots those and proposes a change. It never makes one on its own — every proposal waits on the **Improvements** tab for your yes.
## The four kinds of proposal
| | What it is |
| --------------- | ---------------------------------------------------------------------------- |
| **New skill** | It found a procedure worth keeping. Approving saves it. |
| **Skill patch** | An existing skill needs a correction. Approving edits it. |
| **Memory note** | Something worth remembering about you or the work. Approving saves the note. |
| **Report** | A summary of what it reviewed. Approving marks it read. |
The tab shows a count when something's waiting.
## Your three options
The change is made. A new skill is saved, an edit applied, a note kept.
You can edit the proposal before accepting it. Most useful on a new skill that's close but phrased loosely — fix the wording, then approve.
Nothing changes. Worth doing when the proposal is genuinely wrong, so it stops cluttering the tab.
**Tweak more than you approve outright.** A proposal your coworker wrote is a decent first draft of a skill. Five minutes tightening the description is what turns it into one that actually gets used.
## Skill consolidation
Separate from the above, there's a switch on the **Capabilities** tab called **Skill consolidation**, and it's **off** by default:
> A periodic review proposes merges and cleanups of the skills this worker authored. Every change needs your approval.
With it on, your coworker periodically reviews the skills it has written for itself and proposes tidying them up — merging two that overlap, retiring one that's obsolete.
It's worth switching on for a coworker that's been working for months and has accumulated a lot of skills. There's little point on a new one.
It reviews skills the coworker wrote itself. Skills you wrote by hand are yours, and it doesn't propose rewriting them.
## Retired skills come back
A skill that's been retired isn't gone. There's an archive on the Improvements tab, and you can restore any of them — or delete one permanently once you're sure.
Useful when a consolidation turns out to have merged away something you wanted.
## Common questions
No. Every proposal waits for you. That's the whole design — a coworker that quietly rewrote its own instructions would be impossible to trust.
Nothing breaks. Proposals accumulate and your coworker carries on as it is. It's worth a look every couple of weeks.
Proposals come out of work it was doing anyway. Skill consolidation is a separate periodic pass that costs a little, which is why it's off by default.
Edit the skill directly, or restore an earlier one from the archive. Nothing is one-way.
## Next steps
The bigger picture.
Where memory notes land.
Writing one yourself.
Where the switch lives.