v0.2.0 — Method guard + retry
Capability-metadata release across both packages: the bridge now retries idempotent calls through cascading socket drops, and UnoQTool gains a user-defined JavaScript guard that vets each LLM invocation at the gate.
@raasimpact/arduino-uno-q-bridge@0.2.0
Added
Bridge.callWithOptions(method, params[], opts)— new public entry point taking{ timeoutMs, idempotent }. Whenidempotent: true, in-flight calls that hitConnectionErrorrace the bridge'sreconnectevent against the remainingtimeoutMsbudget and retry — repeatedly, through cascading drop/reconnect cycles, until the call resolves or the budget runs out. Never retries onTimeoutErroror non-idempotent calls. Calls that start against a known-disconnected bridge fast-fail withConnectionErrorinstead of waiting out the timer.
Changed
call()andcallWithTimeout()are now thin wrappers overcallWithOptionswithidempotent: false. Behavior preserved for existing callers.
Full changelog: packages/bridge/CHANGELOG.md
n8n-nodes-uno-q@0.2.0
Added
- UnoQTool — Method Guard field. Optional JavaScript body that runs at invocation time with
methodandparamsin scope and decides whether the call proceeds. Typical uses: validating LLM-supplied arguments, time-of-day gating, external-state checks. Return a string to reject — it's surfaced as structured tool output so the AI Agent feeds the message back to the LLM for self-correction. Uses n8n'sjsEditorwidget. - UnoQTool — Idempotent top-level checkbox driving the bridge's retry on mid-call
ConnectionError.
Changed (minor breaking)
- UnoQCall — Idempotent moved from Options collection to a top-level checkbox for consistency with UnoQTool. Workflows saved with
options.idempotent: truepre-0.2.0 silently reset to defaultfalseon load; re-tick at the top level. No runtime error — just quiet loss of the setting.
Removed
safeReadOnlyboolean (advisory, never enforced). Replaced by Method Guard, which vets each invocation rather than signalling a blanket per-method claim.
Full changelog: packages/n8n-nodes/CHANGELOG.md
All existing unit tests pass (26); bridge integration suite against a real UNO Q also passes (10/10). The method-guard path has been manually verified in a live n8n workflow — guard rejections reach the LLM as observations rather than as workflow errors.