- Home
- Skills
- Workflow Automation
- Durable Task Orchestration
Durable Task Orchestration
Supervise local background jobs, verify outputs, and resume exact agent sessions using a separately installed MIT runtime.
$9
Works with the AI Agents you already use
Durable Task Orchestration
Example session with this skill installed
For an explicitly approved local documentation audit, plan two independent review tasks with separate output files. Use the installed runtime, no new purchases or provider accounts, verify the reports, and acknowledge only their checked results.
- Read your context and instructions
- Compiled the durable task orchestration
Illustrative workflow output, not a customer testimonial or benchmark: task objectives and output ownership recorded; bounded jobs supervised; report artifacts checked; failed or uncertain work identified; exact result records acknowledged only after review. Actual job IDs, reports and provider usage depend on the approved run.
Connects securely to your tools. The creator never sees your data.
About this skill
Plan, supervise, verify
Turn an approved local task into bounded background work with clear output ownership, explicit acceptance checks and durable result records. This product is a workflow instruction layer, not a hosted model or bundled runtime.
What the workflow covers
- Split independent tasks without concurrent edits to shared files.
- Use explicit time and log limits, inspect status and bounded output, and stop only owned work.
- Reconcile partial effects before exact-session follow-up; never blindly rerun an unknown job.
- Verify the requested artifact before acknowledging its result record.
Pinned native workers are read-only by default: they return findings or patch proposals, and the parent writes approved reports. This does not silently enable worker write tools or bypass a sandbox.
Required software and costs
Separately install and review the MIT durable-agent-runtime (Node22+, local POSIX shell): https://github.com/roshameow/durable-agent-runtime/tree/6a9f28834963eb03018c3a1a18188554b3b23cf3 The runtime core requires neither Pi nor RMUX. Optional agent CLIs and their model accounts are separate. This one-time product price does not include provider credits, subscriptions, installation or a managed cloud service.
Compatibility and evidence
macOS is tested; Linux is not independently verified; Windows is unsupported. The runtime has52 offline tests, a narrow live Codex subscription start/exact-session recall test, and a configured Pi-worker coding API smoke. These do not certify this Skill's activation across all agents or guarantee outcomes. Your host needs local shell and appropriate file permissions.
Licence and boundaries
This package contains original instructions only. The external runtime remains independently MIT licensed; its copying and use rights are unchanged. No runtime code, provider key, model credit or wallet capability is bundled. No persistent PTY, remote-machine control, automatic chat notifications, arbitrary crash adoption or exactly-once external actions are promised.
Windows is not supported. Requires manual installation of the durable-agent-runtime binary. Does not support more than four concurrent tasks per state root.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Requires
Permissions
Read/write access is limited by user approval, not an OS sandbox. Runtime creates private job state and an owned temporary Unix control socket. Network/environment access is needed only for approved optional engines, authentication and dependency download. Do not provide browser cookies, credentials, wallet keys or payment authority. No automatic provider purchase, API-billing fallback, publishing or recharge.