Skip to main content

agenta-apps

How to give the person a small interactive page in the drive: a board, checklist, queue, form, table or dashboard. Read it when the ask is for something to look at and touch rather than a reply. Covers picking a starter with create_app, the app folder layout, and the window.agenta bridge for hand-written apps.

Source facts

Repository
Agenta-AI/agenta
Last source activity
October 5, 2026 at 12:36
Detected SKILL.md language
English
Stars
4,887
Forks
726

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
9 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
agenta-apps
description
How to give the person a small interactive page in the drive: a board, checklist, queue, form, table or dashboard. Read it when the ask is for something to look at and touch rather than a reply. Covers picking a starter with create_app, the app folder layout, and the window.agenta bridge for hand-written apps.
# Agenta apps ## When The person asks for something to look at and touch rather than a reply: a board, a checklist, a queue, a form, a table, a small dashboard. Build an app folder in the drive; the drive renders it in a sandboxed page and shows it as a card in the chat. Do not build one for a one-off answer. ## First 1. List `apps/*/app.json`. If an app already covers the ask, update its files; never make a second copy of the same app. 2. Pick a starter from the table below and call `create_app(starter, "apps/<slug>")`. Then write its config file and, when the person gave you content, its data file. 3. Only when no starter fits, write `app.json` and `index.html` yourself (rules below). ## Folder `apps/<slug>/` holds `app.json` (the manifest: `agenta_app: 1`, `name`, `entry`, `access`, `data`, `config`), `index.html`, and its data as sibling JSON files. Set `access` to `read-write` if the app saves anything, else `read`. The app owns what `data` names while it is open; you own config and the rest. You may write data too, but read it first: your writes are unconditional. ## Bridge The page gets `window.agenta`: `await agenta.ready` first; then `canWrite`, `visible`, `dir`, and `agenta.fs` with `read`, `readJSON`, `write`, `writeJSON`, `list`, `exists`, `stat`, `remove`. Paths are relative to `index.html` and cannot leave the folder. Events: `agenta.addEventListener("changed", cb)` when a file changes underneath the app, `"access"` when `canWrite` changes, and `"visibilitychange"`. Failures reject with `error.code`: `not_found`, `read_only`, `scope`, `conflict`, `too_large`, `bad_request`, `unavailable`. Writes carry If-Match automatically. ## Custom app rules - Layout with the kit classes only: `ag-app`, `ag-toolbar`, `ag-btn`, `ag-btn-primary`, `ag-input`, `ag-select`, `ag-check`, `ag-card`, `ag-columns`, `ag-column`, `ag-list`, `ag-grid`, `ag-badge`, `ag-empty`, `ag-toast`. Inline CSS only for what they do not cover. - Keep assets self-contained. Do not assume network isolation or embed secrets. - Wait for `agenta.ready`; treat `not_found` as empty; while `canWrite` is false, show edits as unsaved instead of failing; on `access`, re-render and save them. - Save whole files, debounced, after each change; handle `conflict` by re-read and reapply once. ## Starters | Starter | Use when | Config keys | Data files | Access | |---|---|---|---|---| | `board@1` | Use this when the person wants a kanban, to-do or status board: cards that move between named columns. | title, columns | board.json | read-write | `create_app` copies the starter; `list_starters` shows this table live. ## After Reply with the file card for `apps/<slug>/index.html` and say which access the app asks for (`read` or `read-write`). ## Agent-level layout: `agent-files/.apps/` `agent-files/.apps/` is your record of the apps you have built, across sessions. Do these in order; each rule is a checklist item. 1. **First `create_app` in this agent's life:** write `agent-files/.apps/layout.json` as `{"version": 1, "created_at": "<ISO 8601>"}`. If the file already exists with a different `version`, stop and say so. Do not guess at a migration. 2. **After every app event in this session** (create, update, archive): rewrite `agent-files/.apps/registry/<session_id>.json` whole, as `{"version": 1, "session_id": "<id>", "apps": []}`, one object per app, string keys `slug`, `path`, `template`, `created_at`, `updated_at`, `state`, `notes`. Never write another session's registry. 3. **Before creating an app:** read every file under `agent-files/.apps/registry/` and merge by slug. If the person is asking for something you already built in this session, update it. If you built it in another session, say so and offer to recreate it here from the same starter. When two sessions hold the same slug, keep the one with the later `updated_at`; if those are equal, keep this session's. Never merge two records into one, and never rewrite the other session's file to resolve it: the merge is what you read, not what you write. 4. **Repeated feedback:** when the same feedback arrives about the same starter twice across sessions, add `agent-files/.apps/notes/<YYYY-MM-DD>-<slug>.md` with one paragraph and the session ids. Never edit or delete an existing note. 5. **Scheduled data:** if an app's data is produced on a schedule (a dashboard, a weekly sweep), propose a schedule in Triggers when you have `create_schedule`. Say plainly that a scheduled run writes into its own session, not this one, so the app will not refresh itself here: the person reopens it after a run, or asks you to copy the new data across. Apps live in the session drive; `create_app` cannot write to `agent-files/`, so do not offer to put one there. 6. **Never** put app data files under `.apps/`. **Never** touch `.apps/kit/` unless the person asks for branding.
View on GitHub