agentleFS
alexderzalexderz/agentic-toolbox/skills/cursor-cloud-agents-when/SKILL.md

cursor-cloud-agents-when

Skill5 starsChanged 46 days agoMITView on GitHub

use this when choosing Cursor Cloud Agent or the local grok CLI for coding or heavy multi-step work — Cloud Agent for a remote repo or pull request, the grok CLI for work on this machine under build quota, and the operator's private SDLC repository for artifacts that live there. if Cursor credits are empty, degrade to the grok CLI. do not use on your own initiative, and do not implement with a bot token.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 14 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈14

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep product GitHub repositories separate from the operator's private SDLC repository.

    line 38
  2. Distinctive:

    Use Cursor Cloud Agent when any of these holds:

    line 43
  3. Distinctive:

    If a requested Cloud Agent run fails on usage, credits, quota, or billing, that request also approves the grok CLI for the same goal.

    line 53
  4. Distinctive:

    If Cloud Agent fails because of usage, credits, quota, or billing:

    line 57
  5. Distinctive:

    If a remote pull request was required, produce the patch on this machine.

    line 61
  6. Distinctive:

    If step 3 applies, open or hand off the pull request from an approved host that already has credentials, or give the operator a …

    line 63
  7. Distinctive:

    If the grok CLI is also unavailable, stop and report both blockers.

    line 66
  8. Distinctive:

    Do not retry in a tight loop.

    line 68
  9. Distinctive:

    Do not implement with a bot token, including for a large patch.

    line 69
  10. Distinctive:

    Do not work around the block with cookies or a Windows host.

    line 71
  11. Distinctive:

    If the repository is ambiguous, confirm it with the operator.

    line 76
  12. Distinctive:

    Do not poll. Wait for the run to complete, then resume.

    line 80
  13. Distinctive:

    Follow up with a reply.

    line 81
  14. Distinctive:

    Keep bank, mail, password-store, VPN, and chat secrets out of the prompt.

    line 83

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Cloud Agent vs grok CLI

Choose the worker for coding or heavy multi-step work: Cursor Cloud Agent, the local grok CLI, or both.

Iron law

Builds go to the grok CLI, Cursor Cloud Agents, or both. Do not implement with a bot token; hand the build to one of those workers.

Hand-off

  1. Hand off the whole ticket.
  2. Let the worker own the implementation loop.
  3. verifier checks the final result.

How remote agents and local CLIs fit the process: Workers. A worker's pull request into this repo still goes through intake: Workers.

Default split

WorkUse
Design, review, scratch work, or build-quota research on this machinegrok CLI
A change that must land as a remote branch or pull requestCursor Cloud Agent
Greenfield work with no repository namedCloud Agent new_repo
SDLC artifacts that live in the operator's private SDLC repositoryThat repository, often through a Cloud Agent pull request
Files that exist only on a host with no remote checkoutgrok CLI or shell on that host
Windows, or any host outside the workspaceOff limits unless the operator approved that host: Never

Keep product GitHub repositories separate from the operator's private SDLC repository.

Prefer Cloud Agent when

Use Cursor Cloud Agent when any of these holds:

  • The deliverable is a branch or a pull request on a connected remote repository.
  • The deliverable lands in the operator's private SDLC repository, connected as a remote repository.
  • The work needs Cursor source control or a cloud VM.

Credit degrade

If a requested Cloud Agent run fails on usage, credits, quota, or billing, that request also approves the grok CLI for the same goal. Fall back without a new ask.

If Cloud Agent fails because of usage, credits, quota, or billing:

  1. Say once that Cloud Agent failed on usage, credits, quota, or billing.
  2. Fall back to the grok CLI for the same goal.
  3. If a remote pull request was required, produce the patch on this machine.
  4. If step 3 applies, open or hand off the pull request from an approved host that already has credentials, or give the operator a branch-ready diff.
  5. If the grok CLI is also unavailable, stop and report both blockers.
  • Do not retry in a tight loop. Fall back as above.
  • Do not implement with a bot token, including for a large patch. Use the grok CLI or a branch-ready diff.
  • Do not work around the block with cookies or a Windows host. Stop and report the blocker.

Launch

  1. If the repository is ambiguous, confirm it with the operator.
  2. Launch with a clear goal, the constraints, and what done looks like.
  3. On launch and on reply, prefer Grok 4.7 unless the operator names another model.
  4. Do not poll. Wait for the run to complete, then resume.
  5. Follow up with a reply.
  6. Interrupt a run only when asked.
  7. Keep bank, mail, password-store, VPN, and chat secrets out of the prompt.

For the grok CLI, the run steps are in grok-acp. Load it when the operator opts in to Grok Build over ACP, or when Credit degrade falls back to the grok CLI.

Roles

Role names and jobs: Roles. Tracker writes: Tracker. Host lock: Never.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.