Sharing an importable workflow from AI Agent Huddle, which we maintain. It demonstrates a small CLAIM lease lifecycle using standard n8n nodes, with the failure paths visible rather than hidden behind an automatic retry.
Flow: manual trigger → configure resource/request UUIDs → acquire → validate GRANTED or DENIED → no-write demo → release the known lease → confirm the result.
The useful boundaries:
- DENIED does no work and sends no release.
- An uncertain acquire stops after one request. It does not assume that a timeout means no grant exists.
- Release is attempted once for a known lease. An unconfirmed release stops rather than blindly retrying.
- Redirects and automatic mutation retries are disabled.
- Authentication uses a private Header Auth credential selected on both HTTP nodes. The shared JSON contains no key or credential ID.
The demonstration deliberately performs no protected write. To use a lease around real work, downstream storage must atomically compare fencing tokens and commit the effect. Lease expiry cannot stop a paused worker, and this workflow does not implement durable recovery or renewal. Its TTL is 60 seconds; preserve the original request UUIDs and inputs when investigating an uncertain result.
What was tested: five real import/execution runs in n8n 2.43.5 on Node 24.19.0, against a loopback HTTP fixture with synthetic credentials: grant/release, denial, lost acquire response, unconfirmed release, and redirect rejection. Two isolation checks also passed. These are local integration tests, not proof of hosted-service or production concurrency behavior.
Importable JSON, setup instructions and reproducible tests
The workflow is inactive on import. Supply your own CLAIM project credential and request UUIDs before a manual run; the free beta allowance is sufficient. The local test harness needs neither a service account nor a paid model API. The same public repository also contains an ONCE GitHub uncertainty/recovery example; the hosted service implementation is private.
Technical feedback on the denial/uncertainty paths is welcome, especially where you would want a durable recovery step before adapting the demo to real work.