Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →avenoxai/omp-fleet
Standalone runner + fleet orchestrator for Oh My Pi (`omp`), a coding-agent CLI that reaches your existing OpenAI Codex subscription through a different harness. Does THREE things and always EXECUTES them (never just describes): (1) general code tasks via `omp -p`, (2) cheap high-volume context gathering on the 25x-cheaper model tier, and (3) parallel fan-out — both in-process subagents and multi-process lanes. Defaults locked: provider pinned to your own subscription, never a metered fallback.
Passed 18 of 21 test cases with the skill, and 4 of 21 without it.
and 21 more in the full report
17 fixed5 passed either way3 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.6-flash on 10/2/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
After running several parallel background tasks with built-in isolation, we have leftover worktrees created by omp. Which omp command lists and clears session worktrees in JSON format?
| Checked | Without skill | With skill |
|---|---|---|
| Recommends executing the omp worktree command. | Missed | Passed |
| Includes the --json flag in the command. | Passed | Passed |
The instructions your model reads when the skill loads.
A single self-contained skill for driving Oh My Pi (
omp) from any agent (Claude Code, Cursor, or your own harness). No external control plane required. Drop this file into.claude/skills/omp-fleet/SKILL.md(or your agent's skills dir) and go.Companion to
codex-fleet— same doctrine, different runtime. Read that one for fleet fundamentals (briefs, worktree isolation, lane discipline); this one covers what changes when the lane isompinstead ofcodex exec.Built and battle-tested by Avenox. Share freely.
When invoked you MUST:
omp -p via the Bash tool. Never write instructions for the user to run themselves.run_in_background: true) for anything likely to take >10s.There is no omp exec subcommand. If you're coming from codex exec, the equivalent is -p / --print:
omp -p "your prompt" # correct — non-interactive, process and exit
omp exec "your prompt" # WRONG — no such subcommand
omp installed and authenticated against at least one provider (omp --help, omp usage).omp usage shows every authenticated account and its remaining limit — run it before a fleet; that's your real capacity ceiling.| Setting | Value | When to override |
|---|---|---|
| Provider | your own subscription, pinned | never — see the provider lock below |
| Executor model | openai-codex/gpt-6-astra:high (:medium for routine lanes) | :xhigh/:max only for an explicitly heavy lane; the model replaced gpt-5.6-sol on 2026-09-05 |
| Recon model | openai-codex/gpt-6-luna:medium | switch to the executor model when the lane must decide, not just read; replaced gpt-5.6-luna on 2026-09-26 |
| Config overlay | --config <skill dir>/codex-only.yml | always pass it on unattended lanes |
| Approvals | --approval-mode yolo | drop it if a human is watching and wants prompts |
test cases passed, vs 4 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.