An action is a workflow you save once in a work folder in Treasure AI Studio and rerun on demand — a weekly report to build, a data check to perform, a review to run. Instead of re-typing the same multi-step prompt every time, you start the whole routine with one click of Run on the work folder page — which opens a fresh chat, already named after the action, in which the AI follows the saved instructions — or by typing /<name> inside a chat.
Under the hood, an action is not a separate kind of object: it is a skill whose definition carries one extra marker. The marker adds no new ability — it exists so Treasure AI Studio can recognize the skill as a job you start deliberately and surface it in the UI, instead of treating it as knowledge the AI applies quietly when your request calls for it.
- Signed in to Treasure AI Studio (Getting Started)
- A work folder to hold the action (Work Folders)
A work folder gives you three places to teach the AI how you work, distinguished by when they take effect:
| Mechanism | When It Applies | Stored As |
|---|---|---|
| Instruction | In every chat in the work folder, always | The folder's CLAUDE.md |
| Skill | Automatically, whenever your request falls in its domain — or on demand, when you type /<name> | .claude/skills/<name>/SKILL.md |
| Action | Like a skill — plus it is listed in the UI, where Run starts it with one click | .claude/skills/<name>/SKILL.md with action: true |
Because an action is a skill, everything true of skills stays true: the AI can still apply an action's instructions automatically when a request matches its description. Nor is the /<name> shortcut an action privilege — by default every skill in the work folder can be started that way too, unless its definition opts out with user-invocable: false. The only thing action: true changes is discovery: it lets Treasure AI Studio recognize the skill as something you run on demand, so the UI can treat it individually — listed at the top of the work folder page's Actions and Skills section, with a Run button.
Actions live inside a single work folder and run in chats saved to that folder. They are separate from the skills that reach the AI through plugins and the marketplace — those apply to your chats everywhere but have no Run button. To share a skill beyond one folder, see Sharing Custom Skills.
The easiest way to create an action is to ask the AI, in a chat inside the work folder. When you ask it to save a workflow you want to trigger again — "save this as an action", "make this report repeatable" — it writes the definition file into the folder and tells you the /<name> to run it with. The AI also offers this on its own when it notices a recurring task, such as a second dated copy of the same report.
You can also spell the definition out yourself — dictate the exact content in a chat and the AI writes the file. An action is a directory under the work folder's .claude/skills/, named in lowercase letters, digits, and hyphens, containing a SKILL.md — for example .claude/skills/weekly-report/SKILL.md:
---
name: "Weekly Report"
description: "Build the weekly marketing report"
action: true
---
You are building the weekly marketing report.
1. ...| Field | What It Does |
|---|---|
name | The display name shown in the Actions and Skills list and used as the title of each chat the action starts. The directory name is the identifier you type after /. |
description | One sentence naming the deliverable or the situation the action serves. It is shown in the list and is what the AI matches against when deciding to apply the skill automatically. |
action | true marks the skill as an action. The flag only controls discovery — whether Studio lists the skill with a Run button and can start it from the UI. It adds no new invocation ability; without it, the definition is a plain skill. |
The body below the frontmatter is the instruction the AI follows on each run, written as directions to the AI. Supporting files can sit alongside SKILL.md in the same directory and be referenced by relative path.
There are two ways to start an action, and they behave differently:
- From the work folder page: open the folder, find the action in the Actions and Skills section, and click Run. This starts a new chat in the work folder, and the AI begins working through the action's instructions.
- From a chat in the work folder: type
/followed by the action's directory name and, optionally, your request — for example/weekly-report focus on the June campaign. This runs the action inside the conversation you are in; no new chat is created.
An action can also start itself: give it a schedule and Treasure AI Studio runs it on a cadence you set, opening a new chat each time, without anyone clicking Run.
A chat started with Run is named after the action's display name rather than auto-titled, so runs of the same action share one name in your chat lists. Hover a chat's relative timestamp to see the exact date and time when you need to tell runs apart.
In the sidebar's Recent Chats and in the folder's Chats section, a chat started with Run carries a small ⚡ badge on its chat icon, marking where it came from. The badge marks only Run-started chats: a chat where you ran the action by typing /<name> keeps the plain chat icon in those lists, even though the action ran in it.
The work folder page lists everything under .claude/skills/ in its Actions and Skills section, shown at the folder's root and split by prominence rather than kind:
- Actions are listed directly under the heading, each row showing a ⚡ marker, the display name, the description, a Run button, and an Add schedule button that opens the schedule dialog for that action.
- Plain skills sit in a Skills (n) sub-group below the actions, collapsed by default, each row marked with ✨.
- Broken definitions — skills and actions whose frontmatter failed to parse — are grouped above the actions in a collapsible Broken definitions (n) group.
Clicking any row opens its SKILL.md in the file viewer, so you can review the definition. A definition with broken frontmatter is not hidden: it moves to the Broken definitions group, where its row shows the parse error in place of the description and offers no Run — the flag that decides whether it is an action is part of what failed to parse. Click the row to see the mistake, then fix it by asking the AI in a chat in the work folder.
Because the only difference is the action: true line, converting is a toggle, not a migration. Open the ⋯ menu on any row in the Actions and Skills section and choose Convert to action — or, on an action, Convert to skill. Studio rewrites the frontmatter flag and the row moves to the other group; the instruction body is untouched.
Convert a skill to an action when you find yourself wanting to trigger it — and an action back to a skill when it should stop appearing as a runnable item but keep informing the AI's work.
An action with schedules is the one conversion Treasure AI Studio asks you to confirm. Its menu item reads Convert to skill (has schedules), and choosing it opens a dialog naming how many schedules point at the action.
Converting does not touch the schedules. They stay registered and keep coming due, but every run fails because their target is no longer an action — and after three consecutive failures each schedule turns itself off. The demoted row also loses its automation chip, so those still-live schedules disappear from the Actions and Skills section entirely; they remain visible in Automations.
So convert back, or delete the schedules first, if you want the automation to stop cleanly. Promoting a skill to an action is never guarded — it breaks nothing.
| Issue | Solution |
|---|---|
| Run is disabled on an action | The definition failed to parse — the row shows the error in place of the description. Click the row to see its SKILL.md, then ask the AI in a chat to fix the frontmatter |
| "Failed to start the chat." after clicking Run | The chat could not be created. Reload the work folder page and run the action again; if it keeps failing, click the row to confirm the definition still exists and has action: true |
| An action does not appear in the Actions and Skills section | Return to the folder's root — the section is not shown while you are viewing a subfolder. Then confirm the SKILL.md sits under the work folder's .claude/skills/<name>/ and the frontmatter has action: true. Definitions in other work folders or from plugins are not listed |
| The section says "Failed to load actions and skills" | The listing request failed; nothing is wrong with your definitions. Click Retry in the section. An empty section with no "Loading..." message is a different case — that folder simply has no .claude/skills/ yet |
Typing /<name> sends as plain text | The name must match the action's directory name, not its display name — check it in the row's path on the work folder page |
- Action Schedules — Run an action on a repeating cadence, without starting it
- Skills & Marketplace — What skills are and how the AI applies them automatically
- Work Folders — The workspace an action lives and runs in
- Sharing Custom Skills — Share a skill with your whole account
- Chat History — Where action-started chats appear alongside the rest