Built-in skills and Memory

A skill is a set of instructions an agent loads when a task calls for it. Fleet ships ten. Each one is off until you turn it on, so Fleet adds nothing to your sessions you didn’t ask for.

To switch them on, open Settings → Skills and choose the Built in tab. Each skill has a switch. Sessions you start afterwards get the skills that are on.

Fleet gives its built-in skills to OpenCode 2, OpenCode and Claude Code. See What works where.

Skill	What it does
fleet-code-review	Reviews a change for bugs: the current diff, a branch or a pull request. It checks each finding before reporting it and ranks them by severity. It fixes them or posts them on the pull request only when you ask.
fleet-debug	Finds the root cause of a bug before changing any code. It lists the likely causes, checks each one, and proposes the smallest fix with a test that fails first.
fleet-simplify	Reviews the code changed in a session or branch for reuse, simplification and efficiency, then applies the fixes. It doesn’t hunt for bugs.
fleet-run	Runs the app to see a change working in the real thing, not only in tests. It covers web apps, servers, command-line tools, libraries and terminal apps.
fleet-walkthrough	Walks you through a change as a guided tour in a page beside the chat. It groups edits into chapters. It explains, it doesn’t review.
fleet-plan	Writes an implementation plan as one interactive page beside the chat, with the decisions you must make placed where they apply. You answer in the page and send the answers back.
fleet-mockups	Builds screen mockups, prototypes and before-and-after comparisons as pages beside the chat.
fleet-design	Guides the look of any page the agent builds, so it looks chosen for its subject: colour, type, layout, light and dark themes, and phone width.
fleet-explain	Shows how something works over time as a short animation, up to 30 seconds, beside the chat.
fleet-mod	Makes a Fleet mod for you. It has no switch of its own: it is on while Mods is on in Settings → Features.

The pages these skills draw need Fleet’s tools, which Pi doesn’t get.

On the Built in tab you can:

• Edit a skill to work your way. Your version replaces Fleet’s in new sessions, and Fleet’s updates never overwrite it.
• Open History to see your versions and switch between them.
• Choose Use Fleet’s to go back to Fleet’s version.

If Fleet changes its version of a skill you edited, the skill says Fleet changed it. Sessions keep using yours. Choose See what changed, then Keep mine or Use Fleet’s.

Skills you installed yourself, the catalog and custom sources are on the other tabs: Installed, Catalog and Custom.

Memory lets agents keep notes so a new session starts out knowing what earlier ones learned. It is off by default. Turn it on in Settings → Memory with Remember what agents learn.

It keeps two lists:

• Repository notes: test commands, conventions, your corrections. They reach only that repository’s sessions, worktrees included.
• Machine notes: disk space, memory, crashes, tools that don’t work here. They reach every session on this computer.

New sessions start with both lists. Say “remember that the tests need the --serial flag” in a session, and the agent saves it. Agents also save a note when something fails and they find what works.

• It costs tokens. Every request a session sends carries its notes: about 400 tokens of rules, plus 30 to 50 per note. A repository keeps up to 40 notes and the machine up to 15. Settings shows the running total.
• Nothing is saved quietly. Each note shows in a small notice with Undo, and stays in the conversation.
• You own the notes. They are plain Markdown files in Fleet’s data folder, one per note, never in your repository.
• Learned notes expire. A note an agent learned lasts a week of use, so a lesson whose cause is gone drops out. If the problem returns, the agent learns it again and the note lasts twice as long. Keep pins one for good. Expired notes sit under Expired.

In Settings → Memory you can switch repository, search, Add a note, edit or forget any note, Clear a list, or Clear all memory. Turning Memory off stops agents saving anything and sessions carry nothing extra.

Memory works with OpenCode, OpenCode 2 and Claude Code. Claude Code also keeps its own memory in ~/.claude, so the two can hold the same lesson twice. To keep one memory, turn Claude Code’s own off. Pi isn’t supported yet.

• Workflows: steps can use a skill, such as fleet-debug in Fix a bug.
• Mods: the add-ons fleet-mod writes for you.
• What works where: which harnesses support skills and pages.