agentleFS
himkthimkt/cafleet/.claude/skills/skill-author/SKILL.md

skill-author

Skill56 starsChanged 4 days agoMITView on GitHub

Integrate a CAFleet Director/member team into a skill, with scoped roles, durable broker coordination, Director wake and explicit review gates.

23 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 18 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    Use CAFleet when independent investigation, implementation or review calls for subagents or agent teams.

    line 10
  2. Very specific:

    Run cafleet doctor, then register the calling Director:

    line 42
  3. Very specific:

    Read and ACK the completed report, inspect the summary and present it.

    line 64

“Common” is measured: kept by at least 6 in 10 of the 18 edited versions of this file we could read. “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

Authoring a CAFleet team skill

Use CAFleet when independent investigation, implementation or review calls for subagents or agent teams. A direct local transformation can stay in the current session. A team has a Director and the members selected for its assignment; the host owns continuation.

Phase prerequisites

BeforeReadLocal responsibility
Designing the workflowCore and backend bindingsDefine outcome, role authority, file ownership, artifacts and user gates
Resolving outputsTask-scope resolution, Director write protocol and member BASENormalize the task folder and inherit supplied BASE in members
SpawningSpawn Protocol and Director policyDiagnose environment, register the Director and select members/models
Rendering promptsCanonical skeleton and audit conventionSupply absolute role/core/backend paths, identity, BASE, host prerequisites and lossless role constraints
CoordinationCoordinationUse artifact pointers and paired issue markers
ReceivingDirector wake and backend recipesArm native reception before ready replies and retain one owned receiver until resolution or explicit handoff
Recovery / finishRecovery / ShutdownPreserve context and tear down owned members before the fleet

Members open their authoritative role first, load their own backend and required skills through its supported loader, then use supplied host-rule equivalents for absent optional files. Ready is the first operational broker command; prerequisite shell reads may precede it only when shell is the available reader.

Integration decisions

Normalize research/design inputs to their task folder before resolving BASE. Agent-only artifacts use hidden folders; deliverables use visible paths. Inherit literal BASE or its documented disabled-audit state.

Render a compact role-by-path prompt to a unique .prompts/<member>-<UTC-compact>.md and preserve that pre-spawn audit input. Leave only {fleet_id}, {member_id}, {director_member_id}, {coding_agent} and {database_url} for CLI substitution; double other literal braces. A missing BASE follows its guarded inline transport branch.

Send individually to selected recipients. Members poll after ready, before reporting and on each wake. Ordinary members receive pane previews; the Director follows Director wake. Native shell dispatch returns submission metadata; inspect output and explicitly continue after completion evidence if the host has not resumed the member.

Worked example: summarize-pr

A Summarizer receives a read-only PR diff and produces a 200-word summary with three risk areas. Resolve actual paths/models/IDs for each run; examples use concrete values for readability.

Resolve and bootstrap

Normalize the task folder, resolve BASE and fetch the diff through the host's authorized GitHub workflow. Define a read-only Summarizer role and write its compact prompt under BASE.

Run cafleet doctor, then register the calling Director:

cafleet fleet create --name summarize-pr-1234 --coding-agent claude --json

Suppose this returns fleet 7 and Director 8. Arm native reception through the shared backend recipe and require its ready record before depending on member replies.

Render and spawn

The prompt references the authoritative role, installed CAFleet core/backend references and the member role's BASE section, host-rule source, assignment, literal identity placeholders and startup ready command. Its output scope names the summary artifact; Git and user communication remain Director-owned.

cafleet member create --fleet-id 7 --name summarizer --description "Summarizes a PR diff and three risks" --file /repo/researches/pr-1234/.prompts/summarizer-20260911T100100Z.md --json

Receive that member's ready and ACK it. Dispatch its scope when inputs exist.

Review and revisions

Receive the Summarizer's report under Director wake.

Read and ACK the completed report, inspect the summary and present it. Record revisions as actionable markers at the artifact pointer and send ready (doc). The member revises, removes the marker and sends addressed (doc); keep the owned receiver through the review loop.

Teardown

After the authorized outcome, consume the final inbox, delete fleet 7 and confirm it is absent from the active list. Collect the receiver's fleet-deleted closure and native task exit. Fleet deletion ends each ordinary member as member deletion does, closing its pane, then deregisters the registry while keeping the Director's pane and retaining history. Preserve artifacts and immutable prompt inputs.

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.