omp-coordination-honesty-not-clone
Skill29 starsChanged 7 months agoMITView on GitHub
omp.sh is a terminal coding agent, not a ThumbGate runtime. Steal named-model routes, /new session preservation, saved handoffs, file-disjoint children, and view-only observation onto the checkout lease. Never install omp. Never open a full-control collab link. Slash: /omp-coordination-honesty-not-clone.
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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
- Distinctive: line 77
Write state and next into a handoff file.
- Distinctive: line 79
If someone watches, they get view-only.
- Distinctive: line 80
Run the doctor on any handoff that mentions omp, a model route, or a second writer.
≈ “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.
Files
omp coordination honesty — compare, do not clone
Goal
Fail-closed coordination for whom: agents sharing the ThumbGate checkout — so a handoff names the model, keeps the previous session, and gives each writer disjoint files. omp stays a neighbor. Orchestra OMP is a different system and stays refused.
Constraints
| NEVER | ALWAYS |
curl the omp installer or install @oh-my-pi/pi-coding-agent | npx thumbgate omp-coordination-honesty --json --map-only |
| Model Auto, Fast, default, or balanced | Named id (grok-4.7 or provider/model-id), Fast off |
/clear when the previous session must stay saved | Session new |
| Agents above one, including omp maxConcurrency 32 | Agents one; the parent integrates |
Full-control collab, bearer write token, wss://my.omp.sh | View-only observation, host local |
| Invent a result for an interrupted tool call | Continue completed entries only |
| Overlapping write paths between owners | Disjoint files; children do not inherit the parent transcript |
HARD fail closed. REFUSE the omp binary and any link that can drive the host. Do NOT install omp.
Reference
- https://omp.sh/docs/sessions (
/newvs/clear,/handoff, crash recovery) - https://omp.sh/docs/subagents (no inherited transcript, file-disjoint ownership)
- https://omp.sh/docs/collab (host authoritative, view-only default)
scripts/omp-coordination-honesty.js- Checkout lease:
scripts/session-lease.js(incident 2026-08-11)
Examples (show, don't tell)
Weak: curl the omp installer, set Model: auto, fan out 32 agents, and share a full-control collab link.
Gold:
$ npx thumbgate omp-coordination-honesty --json --map-only
ok: true
$ npx thumbgate omp-coordination-honesty --json --handoff=handoff.md
ok: true # Model grok-4.7, Session new, Agents one, Install none, Host localHandoff header:
Model: grok-4.7
Route: named
Fast: off
Session: new
Think: off
Agents: one
Install: none
Host: localProcedures
npx thumbgate omp-coordination-honesty --json --map-only
npx thumbgate omp-coordination-honesty --json --claim='curl https://omp.sh/install'
npx thumbgate omp-coordination-honesty --json --handoff=handoff.md
npm run test:omp-coordination-honesty- Claim the checkout lease before mutating shared files.
- Write state and next into a handoff file. Start a new session so the old one stays saved.
- Assign each child disjoint paths. The parent integrates. Do not pass the parent transcript.
- If someone watches, they get view-only. Do not open a write link.
- Run the doctor on any handoff that mentions omp, a model route, or a second writer.
Rubric
--map-only→ok=true- installer,
@oh-my-pi/pi-coding-agent, Auto, or Fast → fail - agents above one → fail
- full-control collab or
wss://my.omp.sh→ fail - overlapping owner paths → fail
- interrupted tool entry with a result → fail
/clearon a handoff → fail- named model, session new, disjoint writes, view-only → pass
- reading
https://omp.sh/docs/sessions→ pass - doctor tests PASS
- evidence:
--jsonin the same turn
In this repository98 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .well-known/llms.txtllms.txt
- activesaddler-curriculum-honestySkill
- article-syndication-pipelineSkill
- autonomous-execution-mandateSkill
- aws-strands-harness-not-cloneSkill
- bolt-forge-opt-in-not-cloneSkill
- ci-gha-buildkite-patterns-not-cloneSkill
- cobble-hot-store-compare-not-cloneSkill
- colab-compute-honesty-not-cloneSkill
- comet-session-loggerSkill
- craft-playbook-gateSkill
- cyberstrike-compare-not-cloneSkill
- datadog-llm-obs-compare-not-cloneSkill
- decisive-step-evaluatorSkill
- deeppattern-discipline-honesty-not-cloneSkill
- earnie-third-author-governanceSkill
- embeddinggemma-multimodal-firewallSkill
- fiddler-control-plane-honestySkill
- gitlab-sandbox-allowlist-not-trustSkill
- graphifySkill
- halo-trace-optimize-not-cloneSkill
- harness-evolution-ceilingSkill
- ideabrowser-connectorSkill
- infoq-architect-honesty-not-cloneSkill
- jit-harness-compare-not-cloneSkill
- meko-datapack-honesty-not-cloneSkill
- mintlify-knowledge-honesty-not-cloneSkill
- model-distress-governanceSkill
- multi-harness-rl-invariantsSkill
- nvidia-specdecode-al-doctorSkill
- openui-catalog-compose-honestySkill
- package-manager-honesty-doctorSkill
- qwenwork-policy-group-honesty-not-cloneSkill
- saas-moat-honesty-not-cloneSkill
- salt-code-policy-honesty-not-cloneSkill
- show-meSkill
- siren-mcp-governanceSkill
- skool-headless-readerSkill
- snapchat-abm-playbook-stealSkill
- strands-harness-sentinelSkill
- tailscale-p2p-governance-not-cloneSkill
- thumbgate-board-loopSkill
- thumbgate-daily-discoveriesSkill
- thumbgateSkill
- token-shunt-honesty-not-cloneSkill
- typesafe-typed-questions-not-cloneSkill
- vs-coverage-gap-not-cloneSkill
- vulnerability-pre-action-diodeSkill
- windbg-evidence-diagnosticsSkill
- wisdom-context-artifact-not-cloneSkill
- zero-outbound-spend-they-pay-usSkill
- zvec-grep-compare-not-cloneSkill
- ci-failure-triageSkill
- evidence-first-answerSkill
- fleet-coordinationSkill
- paranoid-mode-honesty-not-cloneSkill
- revenue-truthSkill
- scope-disciplineSkill
- The first 60 shown.
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
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.

