Skills
Reusable instructions for kinds of work. 51 come with Asyncat, and you can save or write your own.
What skills are
A skill is a Markdown file with instructions for one kind of work: how to debug, how to review code, how to set up a Cat. It has a name, a description, a group, tags and a note on when to use it.
Skills don't run anything. For each run, Asyncat picks up to five skills that fit the task, using your model to choose (or matching tags when it can't), and gives them to the agent before it starts. The agent can also look for more with list_skills and load_skill. See Agent tools.
Browse them in Customize → Tools & Skills → Skills. Filter by group, or by where they came from: Built-in, Yours, or Automatic (written on their own by older versions of Asyncat).
Bundled skills
Asyncat ships with 51 skills, in six groups. The full text of each is in the source repository.
| Group | Skills |
|---|---|
| Planning Shaping a goal and reasoning step by step | plan, architecture-review, code-review, rest-api-design, api-endpoint-explorer, codebase-metrics, test-runner, ui-ux-review, design-workspace, data-interpretation, statistics-interpretation, financial-analysis, project-management, stem-math-visualization |
| Workflow Procedures and repeatable ways of working | agentic-coding, systematic-debugging, effective-testing, tdd, refactoring, git-workflow, error-recovery, react-patterns, nextjs-patterns, docker-basics, ci-cd-pipeline, deployment-patterns, database-migrations, sql-queries, data-engineering, performance-optimization, monitoring-setup, log-analysis, cron-jobs, cats, document-generation, effective-documentation, report-writing, browser-utility-artifacts |
| Recall Finding context and what was learned before | onboarding-new-codebase, local-rag, web-research, effective-logging |
| Risk checks Safety, permissions and care with big changes | security-best-practices, dependency-audit, incident-response, error-handling, accessibility-audit |
| Habits Default behaviour and working preferences | None bundled. Your own skills about how you like to work fit here. |
| Tone Style, empathy and how answers read | email-drafting, customer-support, marketing-content, outbound-notifications |
Save this as a skill?
Turn on Suggest skills from my chats in Settings → Privacy & safety → Learning from chats. It starts off.
- After a chat where the agent worked through a task in several steps, one more request to your model drafts a skill from those steps.
- The chat shows Save this as a skill? with the skill's name and description.
- Save writes it to
~/.asyncat/skillsas one of yours. Not now drops it.
Nothing is written until you pick Save. A chat keeps one draft at a time, and turning the switch off drops any that are waiting. Temporary chats, Cats and helper agents never suggest skills. See Learning you control.
Your own skills
On the Skills page, make a new skill with a Name, Description, Category, Tags, When to use and Instructions (Markdown; numbered steps work well). You can delete your own skills there; the bundled ones stay.
Your skills are Markdown files in ~/.asyncat/skills. You can also write one there yourself, then press Reload from the skills folder:
---
name: my-deploy-flow
description: Our team's standard deployment checklist
brain_region: cerebellum
tags: [deployment, production]
when_to_use: When deploying to production or staging
---
1. Run the migrations before deploying.
2. Check the rollback steps before starting the deploy.
3. Post the pull request link in the deploys channel.brain_regionholds the group:prefrontal(Planning),cerebellum(Workflow),hippocampus(Recall),amygdala(Risk checks),basal_ganglia(Habits) orlimbic(Tone). The plain group name works too.description,tagsandwhen_to_usehelp Asyncat pick the skill for the right tasks.- The Skills page won't give a new skill a built-in skill's name. A file you write yourself with a built-in's name takes its place.