agentleFS
phnx-labsphnx-labs/rush-mcp/skills/hand-off-a-task/SKILL.md

hand-off-a-task

Skill0 starsChanged 4 days agoMITView on GitHub

Hand a task to a Rush cloud agent and bring back the result. Use when someone wants work done off their machine or in the background (write tests, upgrade a dependency, research and write a report, investigate an incident), or asks what happened to a task Rush ran earlier.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Start it: Call create_session once, with a fresh UUID you write yourself as request_id (no shell command needed).

    line 28
  2. Very specific:

    Work started earlier: When the person asks about Rush work without an id, call list_sessions (newest first.

    line 51

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Hand off a task to Rush

Rush runs the task in its own cloud computer and keeps going while this conversation moves on. Tool names may carry a client prefix (mcp__rush__create_session). Never ask for a password, token or API key in chat, and never report a result Rush did not return.

  1. Write the task. The session knows nothing from this conversation. Put in task the goal, the files, errors, links and decisions that matter, and what done looks like: "open a pull request", "write the report as report.md", "answer in a few sentences". For code, add "do not merge" unless the person asked for a merge; the agent may otherwise follow merge rules of its own. To build on a finished session, pass its id as continue_from; its answer goes at the top of the new task.
  2. Pick where it runs. For code, pass repo: a GitHub repository, branch (/tree/<branch>) or pull request URL from list_repositories, or a gitlab.com project. If the team keeps a Project for that repository (list_projects), pass project instead; it brings the Project's standing instructions and secrets. Name a saved agent with agent when one fits (list_agents), and the accounts the task may use in connections (list_connections). Leave out what the task does not need; with no agent, Claude Code does the work.
  3. Start it. Call create_session once, with a fresh UUID you write yourself as request_id (no shell command needed). If the call errors and you retry, send the same request_id so only one session starts. For independent tasks, start one session each.
  4. Wait. Tell the person it is running, then call wait_for_sessions with every id and timeout_seconds: 50, again until settled is true. A session past the runs-at-once limit starts queued; keep waiting. With several, until: "any" lets you report each one as it lands.
  5. Read the result. get_session has the answer, branch, pull request and the files it made. get_session_changes lists the changed files (include_diff for the diff); get_session_artifact reads one file by name. Use list_session_events only to explain how the agent got there or why it failed.
  6. Continue or stop. send_session_message continues in the same computer, with its files and conversation, for 7 days after it goes idle; follow it with wait_for_sessions. cancel_session stops the session and releases its computer, after which no follow-up is possible; ask before stopping one.

Report each session in a few lines: its status, the answer, and the pull request or file links. needs_review means Rush could not confirm the result: say so and show what the agent reported. On failed, give the error and the next step; do not retry silently.

Work started earlier. When the person asks about Rush work without an id, call list_sessions (newest first; filter by status), match on the first line of each answer, then read the one they mean with get_session.

When Rush refuses

A refused call's text starts with a code, and often a Next: step; follow it. For these, tell the person the fix in one line:

CodeFix
TOKEN_LIMIT_REACHED, CREDITS_EXHAUSTED, SPEND_CAP_REACHED, PAYMENT_METHOD_REQUIRED, COMPUTE_LIMIT_REACHED, WEEKLY_COMPUTE_LIMIT_REACHEDRelay the message; it names the fix. Model accounts are at https://cloud.getrush.ai/console/settings/?section=accounts and billing at https://cloud.getrush.ai/console/settings/?section=billing
RESOURCE_NOT_ACCESSIBLERush's GitHub App does not reach that repository. Add it at https://cloud.getrush.ai/console/settings/?section=connections, or run without repo.
CONNECTION_NOT_READYCall connect_account with type app (gmail, linear, slack, …) or mcp (with its url), and name; for an agent made from a template, also agent. A gitlab.com project needs name gitlab. Give the authorize_url to the person to open in their own browser; never open or fetch it. It works once, for 10 minutes. Start the session again once list_connections shows it connected.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.