Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Dynamic multi-agent workflows — plan first, then orchestrate parallel agents with adversarial verification via the local odw daemon. Use when the user asks for a "workflow", says "ultracode", or hands you a task spanning many files/items that benefits from parallel agents.
.claude/skills/bilal140202-odw/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -26% | 0% |
You orchestrate large tasks through an explicit plan and a local daemon instead of improvising. The scripts referenced below live next to this skill in scripts/.
Codex exposes no API for an extension to invoke its own model, so — unlike the OpenCode plugin, which runs ODW's real engine through OpenCode's configured model with no extra key — there are exactly two honest paths here:
~/.odw/config.json (Anthropic / OpenAI-compatible / Ollama — Ollama is keyless/local). This is the only way to get the full engine on Codex today.There is no middle option unless/until Codex ships a host-model API or MCP sampling. Pick the path based on whether the daemon is up (Step 0).
Run: node scripts/daemon-bridge.js --check
npm install, npm run setup, then odw-daemon start).node scripts/daemon-bridge.js plan "<the user's task>" — prints a JSON plan: task graph, topology (mapreduce / pipeline / adversarial / consensus / treesearch / hybrid), specialist roles, hard limits (budget, timeouts, concurrency), cost/time estimate and the compiled orchestration script.node scripts/daemon-bridge.js exec plan.json — returns a wf_... id. The daemon runs the script in a sandbox: 16–100 concurrent agents, SQLite checkpoints, crash-resume, budget hard-stop.node scripts/daemon-bridge.js status <wf_id> while running; node scripts/daemon-bridge.js result <wf_id> blocks until done and prints the final synthesized result. Relay it to the user.Orchestrate with your own subagent capability, platform-limited:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 13,448 | 5,397 | -60% | 1 | 1 | 0% | 2,425 | 1,843 | -24% | 0 | 0 | — |
case-01 | fail→fail | 8,073 | 13,873 | +72% | 1 | 1 | 0% | 1,064 | 2,759 | +159% | 0 | 0 | — |
case-02 | fail→fail | 14,146 | 4,835 | -66% | 1 | 1 | 0% | 2,388 | 1,098 | -54% | 0 | 0 | — |
case-03 | fail→fail | 19,418 | 7,336 | -62% | 1 | 1 | 0% | 3,715 | 1,037 | -72% | 0 | 0 | — |
case-05 | fail→pass | 11,677 | 2,155 | -82% | 1 | 1 | 0% | 2,187 | 1,157 | -47% | 0 | 0 | — |
case-06 | fail→pass | 8,847 | 1,943 | -78% | 1 | 1 | 0% | 1,625 | 1,153 | -29% | 0 | 0 | — |
case-07 | fail→pass | 10,226 | 1,554 | -85% | 1 | 1 | 0% | 1,868 | 1,041 | -44% | 0 | 0 | — |
case-08 | fail→pass | 14,808 | 6,595 | -55% | 1 | 1 | 0% | 2,697 | 1,993 | -26% | 0 | 0 | — |
case-13 | pass→pass | 14,000 | 8,142 | -42% | 1 | 1 | 0% | 2,675 | 2,257 | -16% | 0 | 0 | — |
case-09 | fail→pass | 12,664 | 1,809 | -86% | 1 | 1 | 0% | 2,482 | 1,054 | -58% | 0 | 0 | — |
case-10 | fail→pass | 7,221 | 1,992 | -72% | 1 | 1 | 0% | 1,173 | 1,147 | -2% | 0 | 0 | — |
case-11 | pass→pass | 5,985 | 3,077 | -49% | 1 | 1 | 0% | 988 | 1,248 | +26% | 0 | 0 | — |
case-12 | pass→pass | 11,209 | 8,592 | -23% | 1 | 1 | 0% | 1,923 | 2,080 | +8% | 0 | 0 | — |
case-14 | fail→pass | 10,602 | 2,468 | -77% | 1 | 1 | 0% | 1,785 | 1,207 | -32% | 0 | 0 | — |
case-15 | pass→pass | 7,428 | 1,942 | -74% | 1 | 1 | 0% | 1,196 | 1,052 | -12% | 0 | 0 | — |
case-16 | pass→pass | 5,734 | 3,222 | -44% | 1 | 1 | 0% | 928 | 1,342 | +45% | 0 | 0 | — |
case-17 | fail→pass | 18,592 | 1,747 | -91% | 1 | 1 | 0% | 1,001 | 1,039 | +4% | 0 | 0 | — |
case-18 | fail→pass | 11,367 | 3,670 | -68% | 1 | 1 | 0% | 1,949 | 1,457 | -25% | 0 | 0 | — |
case-19 | pass→pass | 10,211 | 3,334 | -67% | 1 | 1 | 0% | 1,681 | 1,287 | -23% | 0 | 0 | — |
case-20 | pass→fail | 4,537 | 6,677 | +47% | 1 | 1 | 0% | 855 | 1,348 | +58% | 0 | 0 | — |
case-21 | pass→fail | 12,384 | 8,943 | -28% | 1 | 1 | 0% | 2,251 | 1,724 | -23% | 0 | 0 | — |
case-22 | pass→fail | 9,933 | 5,125 | -48% | 1 | 1 | 0% | 1,928 | 970 | -50% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +32 percentage points is the difference between those two pass rates over the 18 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.