First Task

First Task Skill

Runs the first chat after setup. This message holds the user's ask, then "What setup learned about me" (their picks and the machine basics), then these rules and a JSON block. Goal: visible work within one minute, a finished, useful result within five.

When to Use

• This message carries the skill under a handoff from setup.
• Until the first result lands. After that it is a normal chat.

Prerequisites

Tools: manage_connections, manage_catalog, clarify, terminal, the file tools, the browser, tool_search, skill_view, and desktop_preview in the desktop app. A tool named here but missing from your tool list, often manage_catalog, is deferred: run it through tool_call, never by its bare name.

How to Run

Obey these limits. They win over every other line.

1. Your first reply is one short line of text and a tool call, in the same reply.
2. The JSON block holds connect (connector ids) and install (plugin ids). The ids are exact. Run the forms below directly, with no search, describe or status step first.
3. A check is any read-only command (ls, which, find, cat, a probe) or a file read. At most 2 checks before you make something. Put several checks in one command.
4. Never search the whole disk (find /) and never read files outside the task's folder. "What setup learned" is the survey.
5. At most one question before work, and only for a vague ask.
6. Never run the same call twice. Change a failed call once. If it fails again, say so and go on.
7. Every first slice makes a thing: a file, a page in the preview, an install, or a brief from real data. A chat summary of what you looked at is not a result.
8. Every first slice ends with the close card, copied exactly from the Quick Reference, also after a failure.
9. Say plainly what failed and what you did not test. Never mock or invent data.

Quick Reference

manage_connections  {"action":"connect","connectors":["<id>", ...]}
manage_catalog      {"action":"install","items":[{"kind":"plugin","id":"<id>"}, ...],"reason":"<one line>"}
tool_call           {"calls":[{"name":"manage_catalog","arguments":{"action":"install","items":[{"kind":"plugin","id":"<id>"}],"reason":"<one line>"}}]}   (when manage_catalog is deferred)
clarify             {"questions":[{"question":"<short question>","choices":["<option>","<option>","<option>"]}]}
close card          {"questions":[{"question":"How does this look?","choices":["Looks right","Change something","Take it further"]}]}

Put one entry in calls for each tool_call. Copy the close card exactly: never rename, add or mark a choice. Your next-step idea goes in the text line.

Procedure

1. Connect first

Run steps 2 and 3 back to back, right after your first line. Skip a step whose list is empty.

Each step runs once. Never redo an earlier step. The connect card's answer is final for this chat: use the apps that connected, and never connect or ask about the others again. An app left unconnected after Continue counts as skipped. Say once, at the close, that they can connect the rest later from the Connections menu.

2. First move for each ask

Options for a vague ask. Each one makes a thing in five minutes. Name only apps they picked. Never offer "find", "review", "audit" or "clean up my <app>": these make a survey, not a thing.

• Work apps picked: "A daily brief from Linear and Slack", "A summary of my week".
• An NVIDIA or Spark machine: "Install a few apps for this <machine>", with the word setup's "This …:" line uses for it (Spark, PC, Mac).
• A plugin picked: "A simple scene in Blender".
• Otherwise: "A small HTML page about <something they picked or said>", "A start page with links to my apps", "A quick script that tidies my Downloads".

The first option always makes a page or a file. Text they type instead of a choice is the pick. If they answer "surprise me", "idk" or "any", start the first option now: write the file and open it, with no checks first.

3. When a connect or an install fails

Do not stop with nothing.

1. No app connected, for a brief: say so in one line. Run one check, gh auth status. If it is logged in, write a brief from GitHub (open PRs, review requests, assigned issues) and say it comes from GitHub. If not, make a start page or a script that needs no account.
2. An install whose state is not connected, or that errors: say so in one line. Do not look for the plugin on disk, read its source or write your own client. Make the nearest thing without it. For Blender: write ~/hermes-first-task/first_scene.py, a script that builds the scene, and give the command blender --python <path>.
3. End with the close card. Offer to try again in the text line.