Open Dynamic Workflows
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/.
Model & API key (read this first)
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:
- No-key path (Native fallback): you, Codex, orchestrate the task with your own model and native subagents (platform-capped, ~6 concurrent). This uses no extra key, but it is not the ODW engine — no sandbox, no 100-way fan-out, no crash-resume, no budget hard-stop.
- Full-engine path (Daemon): the real ODW engine runs in the local daemon, which uses its own provider key in
~/.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).
Step 0 — Daemon check
Run: node scripts/daemon-bridge.js --check
- Exit 0 → daemon is up; use the daemon path.
- Exit 1 → daemon is down; use the native fallback path (and mention once that the daemon installs from github.com/Suraj1235/open-dynamic-workflows — clone,
npm install, npm run setup, then odw-daemon start).
Daemon path
- Plan:
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.
- Confirm: show the user a compact summary (topology, agent count, est. cost, est. time) and ask before executing anything beyond read-only work.
- Execute:
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.
- Monitor / report:
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.
Native fallback path (no daemon)
Orchestrate with your own subagent capability, platform-limited:
- Decompose into discovery → parallel work → adversarial verification → synthesis. State the plan first.
- Fan out independent items to subagents with hyper-scoped instructions and structured JSON outputs.
- Verify aggregated results (hunt false positives, challenge severity, find gaps) before synthesizing.
- Respect safety: never write files or run mutating commands without explicit user approval.
Safety rules (both paths)
- Read-only operations are auto-approved; writes, shell commands and git commits need user approval.
- Respect the budget: if the plan estimate looks expensive, say so before executing.
- Never put secrets in prompts, plans, or logs.
1---2name: odw3description: 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.4---56# Open Dynamic Workflows78You 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/`.910## Model & API key (read this first)1112Codex 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:1314- **No-key path (Native fallback):** *you, Codex,* orchestrate the task with your own model and native subagents (platform-capped, ~6 concurrent). This uses no extra key, but it is **not** the ODW engine — no sandbox, no 100-way fan-out, no crash-resume, no budget hard-stop.15- **Full-engine path (Daemon):** the real ODW engine runs in the local daemon, which uses **its own** provider key in `~/.odw/config.json` (Anthropic / OpenAI-compatible / Ollama — Ollama is keyless/local). This is the only way to get the full engine on Codex today.1617There 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).1819## Step 0 — Daemon check2021Run: `node scripts/daemon-bridge.js --check`22- Exit 0 → daemon is up; use the daemon path.23- Exit 1 → daemon is down; use the native fallback path (and mention once that the daemon installs from github.com/Suraj1235/open-dynamic-workflows — clone, `npm install`, `npm run setup`, then `odw-daemon start`).2425## Daemon path26271. **Plan:** `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.282. **Confirm:** show the user a compact summary (topology, agent count, est. cost, est. time) and ask before executing anything beyond read-only work.293. **Execute:** `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.304. **Monitor / report:** `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.3132## Native fallback path (no daemon)3334Orchestrate with your own subagent capability, platform-limited:351. Decompose into discovery → parallel work → adversarial verification → synthesis. State the plan first.362. Fan out independent items to subagents with hyper-scoped instructions and structured JSON outputs.373. Verify aggregated results (hunt false positives, challenge severity, find gaps) before synthesizing.384. Respect safety: never write files or run mutating commands without explicit user approval.3940## Safety rules (both paths)4142- Read-only operations are auto-approved; writes, shell commands and git commits need user approval.43- Respect the budget: if the plan estimate looks expensive, say so before executing.44- Never put secrets in prompts, plans, or logs.
hashgraph-online