agentleFS
woodbearywoodbeary/later-dog/.agents/skills/orchestrate-cloud-work/SKILL.md

orchestrate-cloud-work

Skill2 starsChanged 29 days agoApache-2.0View on GitHub

Run a sustained program of repository work through later.dog's supervisor: split it into scoped cloud jobs with dependencies, follow their durable receipts, and carry each result to a verified, reviewed draft PR.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is 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.

  • 1

    Very specific

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

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

    Submission uncertain: never submit again. Look for the task with list_provider_tasks on the job's profile and attach it with …

    line 23

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Orchestrate cloud work

Use when the work is bigger than one change: several jobs, several repositories, or days of follow-up. Coding runs in the cloud through the supervisor; do not fan out parallel agents on the person's computer instead. Use the later.dog MCP server. Discover tools by the names below; client-specific prefixes may vary.

Plan

  1. Read the workspace. laterdog_workspace lists repositories, account profiles, backends and their stated limits, slots in use, existing jobs and measured outcomes. inspect_provider_access shows a profile's sign-in and CLI capabilities, not its remaining allowance; treat allowance as unknown. A backend listed as enabled is ready to use: later.dog Cloud runs each job in its own cloud computer on the owner's Cloudflare with their Claude plan, and needs no keys from the person. Say a key or setup step is missing only when the workspace shows that backend disabled, and quote its stated reason.
  2. Split the work by what each piece owns. Every job gets a goal, acceptance criteria, write scopes as narrow as the change allows, and dependencies on the jobs it builds on. The supervisor holds a job until its dependencies are ready and runs overlapping scopes in one repository one at a time; choosing good boundaries is still your call.

Delegate and follow

  1. delegate_cloud_job for each job, with a requestKey tied to that intent (reuse it on a retry, never for different work), the repository, a title, a brief holding the goal and acceptance criteria, writeScopes, dependencies, and the program's standingInstructions on every job. Leave profileId out to use later.dog Cloud; name a profile only to send the job to a different backend the workspace lists as enabled. The person sees each job as a cloud task card in the chat, with its live progress, so tell them what you started rather than polling it. A brief goes to the cloud: no credentials or private material in it.
  2. inspect_cloud_job reads a job and its receipts without touching the provider task. Never resume a task just to see whether it is alive.
  3. A finished job is a result to judge, not proof. Read the diff with collect_cloud_diff, publish it as a draft PR with publish_cloud_job (only the remote publishing host can; a supervisor on the local computer refuses), then follow the monitor-and-ship skill. Merging stays with the person.

When something goes wrong

  1. Submission uncertain: never submit again. Look for the task with list_provider_tasks on the job's profile and attach it with reconcile_cloud_task. Only if that list shows no task may cancel_cloud_job with acknowledgeUnknown free the slot.
  2. Cancelling stops queued work. A native task that was already submitted cannot be cancelled; the job records the request and the task may keep running.
  3. Limits stay visible. This adapter cannot create a cloud environment, continue a native task or cancel a submitted one. A follow-up change is a new repair run (correct_cloud_job) on the published branch.

Keep score

  1. Record decisions, human interventions, regressions and sourced costs with record_job_observation. Report PRs opened, verified and merged as separate numbers, with blockers and limits. A capacity goal (a monthly PR count, a concurrency level) is a target until observations show it; raise concurrency only after they do.

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.