| id | 7953ed01-9be0-5756-b063-39298f6c5186 |
| name | flowpad-assistance |
| description | Flowpad assistance — drives the Flowpad app on behalf of the user. Supported actions:
• context — read what the user is currently looking at (current project,
process, workspace, etc.) so other actions can compose without asking
for an id. Triggered by phrases like "current X", "this X", "the
one I'm in".
• navigate — drive the active browser tab to a Flowpad entity by TypeId
(e.g. "shell-<uuid>", "markdown-<uuid>", "project-<uuid>",
"agentic_process-<uuid>"). Triggered when the user asks to open, show,
navigate to, or jump to an entity. See ``navigate.md``.
• records — entities/records CRUD. Discover types via ``flow schema list``,
get the JSON shape via ``flow schema info <type>``, materialize a
record on disk, then run ``flow record index <path>`` to persist it.
Triggered by "create a task / skill / agent / workflow", "make a new
X", "add an X". See ``records.md``.
• search — Search local files, skills, agents, specs, plans and other
assets via ``flow record search <query> <time> <limit>`` (SQLite FTS5).
Use this whenever the user asks to find / look up / show / search for
something without giving an explicit TypeId. See ``search.md``.
• process — restart the calling agentic-process session via ``flow process restart``
(kills + re-spawns the worker, resuming the session). Use after installing an
MCP server so the new config loads, or when the user says "restart this
session / process / agent". Defaults to the current process. See ``process.md``.
• message — send a message (optionally with file attachments) into a
conversation via the backend's ``add_message`` HTTP action (there is no
``flow`` CLI command for this). Triggered by "send X to my conversation
with Y", "attach this doc to the Z conversation", "message Y the report".
See ``message.md``.
NOT handled here: building a website / web app / SaaS / dashboard — even when
phrased as "using flowpad assistant". That is the separate ``web-app-builder``
skill (copy-as-is template + setup); invoke it instead of the records action.
|
| tags | ["flowpad","ui","navigation","context","records","search","message"] |
| allowed-tools | ["Bash(flow context list:*)","Bash(flow navigate entity:*)","Bash(flow schema list:*)","Bash(flow schema info:*)","Bash(flow record index:*)","Bash(flow record search:*)","Bash(flow process restart:*)","Bash(curl:*)","Read","Write","Edit"] |
Flowpad assistance
A multi-action skill for the Flowpad app. Identify the requested action from the user message, then follow the matching section/file below. Actions can be composed: a request like "create a task and open it" is records (creating) followed by navigate (opening).
Building a website or web app is not an action of this skill. If the
user asks to build/create a website, web app, SaaS, dashboard, or landing
page — even phrased as "build me a website using flowpad assistant" — stop
and invoke the web-app-builder skill, which bootstraps a tested full-stack
template into the session's working directory. Do not hand-write HTML/JS
files or route this through records.
Action: context
Read the per-tab data-context the UI mirrors to the server. Use this whenever the user refers to "the current X" / "this X" / "the one I'm looking at" instead of giving an explicit TypeId.
How to read context
flow context list
That is the entire invocation. No flags needed for the active tab. Pass --connection-id <id> only if the user explicitly names a different connection.
Output
Success — exit 0, one JSON line on stdout:
{
"ok": true,
"connection_id": "...",
"context": {
"CurrentProjectTypeId": "project-<uuid>",
"CurrentProcessTypeId": "agentic_process-<uuid>",
"CurrentWorkspaceTypeId": "workspace-<uuid>",
"CurrentComputeNodeTypeId": "compute_node-@local",
"CurrentUserTypeId": "user-<uuid>",
"CurrentActiveEntityTypeId": null,
"CurrentDomainTypeId": "...",
"CurrentVisitorTypeId": null,
"CurrentAgentTypeId": null,
"CurrentWorkflowTypeId": null,
"CurrentFlowTypeId": null
}
}
Each value is a TypeId string (<type>-<id>) or null if not set. The exit codes match other actions: 3 no active tab, 4 connection not found, 5 server unreachable.
Mapping user phrases → keys
| User phrase | Read this key |
|---|
| "current project" / "this project" | CurrentProjectTypeId |
| "current process" / "this agent" / "this session" | CurrentProcessTypeId |
| "current workspace" | CurrentWorkspaceTypeId |
| "current compute node" / "this machine" | CurrentComputeNodeTypeId |
| "current user" / "me" | CurrentUserTypeId |
| "current agent" | CurrentAgentTypeId |
| "current workflow" | CurrentWorkflowTypeId |
| "current flow" | CurrentFlowTypeId |
| "active entity" / "what I'm focused on" | CurrentActiveEntityTypeId |
If the matched key is null, tell the user that nothing is set for that scope — do not guess or fall back to a different entity.
Other actions
For any action other than context, open the matching file in this skill directory and follow it literally. Do not improvise from the action summaries above — those are routing hints, not specifications.
When composing actions (e.g. "navigate to the current X" or "create a task and open it"), read both files end-to-end before you start running commands.
