grok-bot
Chat with Grok Bot teammates, message a named Grok Bot, create a new Grok Bot, list Grok Bots, read a Grok Bot transcript, or operate the xAI Grok Bot desktop app from Cursor, Claude Code, or Codex instead of the UI. Use when the user mentions Grok Bot, Grok teammates, Reed, Agent Computer, spinning up a new bot, or talking to Grok from the terminal.
How do I install this agent skill?
npx skills add https://github.com/adamanz/grok-bot-skill --skill grok-botIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The Grok Bot skill provides a command-line interface to interact with the xAI Grok Bot desktop application. It securely handles local session tokens by retrieving them from the macOS Keychain and decrypting them using local utilities. The skill communicates with the official service backend to manage bot teammates and chat history. It includes an inherent risk of indirect prompt injection due to its ability to read external chat transcripts, though it implements best practices to prevent secret leakage.
- Socketpass
No alerts
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
Grok Bot
Talk to existing Grok Bot teammates and create new ones through the local CLI. Do not drive the Grok Bot desktop UI and do not call the gateway by hand.
Workflow
- Run
status, thenlist. - Message an existing bot with
chat(waits for a reply) orsend(fire-and-forget). - Create a new bot with
createonly when the user asked for a new teammate. - Prefer one focused job per bot. Put standing rules in
--description. Keep first tasks draft-only.
The CLI is scripts/grokbot.py next to this file. Resolve that path from the
installed skill directory (often ~/.cursor/skills/grok-bot/scripts/grokbot.py).
# After a global Cursor install. For other installs, use the absolute
# path to scripts/grokbot.py inside the directory containing this SKILL.md.
SCRIPT="$HOME/.cursor/skills/grok-bot/scripts/grokbot.py"
python3 "$SCRIPT" status
python3 "$SCRIPT" list
python3 "$SCRIPT" chat --name Reed --prompt "What are you waiting on from me?"
python3 "$SCRIPT" create --name Reed --title "Chief of staff" --description "Draft-only daily brief. Never send messages."
python3 "$SCRIPT" transcript --name Reed --limit 20
Output is JSON. Summarize it for the user. Never paste tokens, gateway URLs, or raw keychain material into chat.
Create rules
A new bot needs:
--name— short, unique--title— one job, not "general helper"--description— outcome, sources, and what it must not do without approval
Do not create a second bot with the same name unless the user asks. The CLI refuses duplicates unless --force is passed.
After create, send one concrete first task with chat unless the user only wanted the empty teammate.
Chat rules
- Resolve bots by
--namewhen unique; use--idif names collide. - Use
chatwhen the user expects an answer in this conversation. - Use
sendfor long-running work the user will check in the Grok Bot app. sendandchataccept--message-id UUIDand returnmessageId. To retry the same message to the same bot after uncertain delivery, reuse that UUID. Network errors during delivery include the UUID. Generate and retain one before sending if the process might be interrupted. Use a new UUID for a new message.- If
chatfails while waiting after sending, checktranscriptbefore retrying. If no message ID was retained, do not blindly resend: a missing transcript entry does not prove delivery failed. - If
stillRunningis true, say so and offer to poll again. Do not invent a reply. - If
timedOutis true, completion is unconfirmed even whenstillRunningis false. ChecknewEntriesortranscript: a queued message may not have started, or a fast reply may have finished between polls. Do not resend with a new message ID. - Do not send passwords, API keys, or 2FA codes. Tell the user to take over Agent Computer in the Grok Bot app.
- Do not approve local-computer access unless the user explicitly asked for it.
Safety
- The CLI decrypts the Grok Bot session from Keychain and
sand-secrets.json. Treat that as a secret broker. Never print or log tokens. - All of a user's bots share one cloud computer. Do not treat separate bots as a security boundary.
- This is not the Cursor Cloud Agents API (
api.cursor.com/v1/agents). Do not create a coding cloud agent when the user asked for a Grok Bot.
Troubleshooting
| Symptom | What to do |
|---|---|
| No saved session | Open Grok Bot and sign in with Cursor. If already signed in, fully quit and reopen the app so it saves the session |
| Unexpected token format / no active account | Update this skill, then select the signed-in account in Grok Bot. The CLI supports legacy scoped tokens and the active entry in the newer saved-account store. Never request the contents of sand-secrets.json or a Keychain dump |
| Keychain read failed | Ask them to unlock the login keychain and retry |
| Access not granted | Eligible plans: SuperGrok Heavy, Cursor Ultra, Cursor Teams Premium |
| Computer still starting | Wait and rerun status until box is running |
| No bot with that name | list and confirm the name |
If the CLI is missing a command the user needs, run python3 "$SCRIPT" --help. Do not invent new gateway routes in chat.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/adamanz/grok-bot-skill/grok-bot">View grok-bot on skillZs</a>