agentleFS
IgorGanapolskyIgorGanapolsky/ThumbGate/.claude/skills/fleet-coordination/SKILL.md

fleet-coordination

Skill29 starsChanged 7 months agoMITView on GitHub

Mandatory coordination contract for ANY agent working in IgorGanapolsky/ThumbGate. Runs automatically at session start (SessionStart hook). Before mutating shared repository or vault state, verify no other live agent owns this checkout, has an active vault claim on ThumbGate, or holds a Linear issue covering the same work. Invoke whenever you are about to edit shared files, create branches/worktrees, push, open a PR, or touch ~/Documents/AI-Agent-Sync. Prevents same-repo collisions (files deleted under a live agent, claims overwritten, work lost).

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:

    Before ANY state mutation (git checkout/clean/reset, branch/worktree creation, push, PR open/merge, vault write), complete these checks:

    line 23
  2. Very specific:

    Never write to Agent-State/<other-agent>.md.

    line 70

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Fleet Coordination — ThumbGate

ThumbGate is worked by several agents (Claude Code, Codex, grok, Gemini, Hermes) fed from the same broadcast prompt, on the same checkout host. Coordination is not optional: the repo ruleset and single-writer lease exist because two agents once mutated the same checkout concurrently and wiped each other's work (incident 2026-08-11, see AGENTS.md "Single-Writer Checkout Lease").

The contract

Before ANY state mutation (git checkout/clean/reset, branch/worktree creation, push, PR open/merge, vault write), complete these checks:

  1. Checkout lease — only one live session owns a checkout. Claim: THUMBGATE_SESSION_AGENT=<session-id> node scripts/session-lease.js claim Verify: node scripts/session-lease.js check Exit 1 = another live agent owns it. Use git worktree add instead.
  2. Vault claims — ~/Documents/AI-Agent-Sync/Agent-Jobs/running/*.md touching ThumbGate belong to other agents unless stale (>7d, dead PID).
  3. Vault dirty state — 180+ uncommitted files is normal; never git add -A there. Stage only your own file.
  4. Linear ownership — issues in In Progress/In Review with a ThumbGate title are claimed. Do not start overlapping work; claim your own issue first.
  5. Herdr live panes — when the gateway is up, same-cwd agents are live collision risk.
  6. Open PR census — never duplicate an open PR or its review threads.

The one-command sweep

bash .claude/scripts/session-bootstrap/fleet-coordination-check.sh

Prints every surface above in one pass (read-only). SessionStart hook already runs it; if you did not see its output this session, run it before the first mutation.

Write-back rules (your own file only)

  • Vault state: Agent-State/<agent>.md may be shared by every instance of that agent kind — check its mtime first. Prefer a dated file under Handoffs/<date>-<topic>.md when in doubt. Commit only your file with git add <your-file> and open a PR (branch protection applies there too).
  • Linear: claim via the Linear bridge or API before long-running work so another agent can see ownership.
  • This repo: never git add -A in a shared checkout. Stage explicit paths.

Failure mode this prevents

Two agents, same repo, both "helpful": one force-checkouts main and cleans, wiping the other's untracked work; the other git add -As the debris into the wrong branch. The lease + claim + census above make that collision visible before it happens.

Concurrent Agent-State writes (HARD)

Never write to Agent-State/<other-agent>.md. Each session appends only its own Agent-State/<this-agent>.md (or Handoffs/*<this-agent>*). Two same-kind sessions must use distinct filenames (e.g. grok.md vs grok-<session>.md) or an atomic claim/lease before overwrite.

Session handoff (omp.sh, do not install)

A handoff is a saved state-and-next file. Starting a new session keeps the previous one. A child does not inherit the parent transcript and writes files the parent does not write. Observation is view-only. Do not install omp or open a link that can prompt or interrupt the host.

npx thumbgate omp-coordination-honesty --json --map-only

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.