Check if execution is allowed for a given service.

* Create a decision point when a task has failed too many times (loop detection).

* Check all active sessions and mark any with dead PIDs as failed. * Called periodically (every minute) to proactively free up concurrency slots

Create a text post on LinkedIn.

Display subject — strip "Re:" but preserve original casing from root message

* Ensure a reply was posted to inbox.json after the agent session completes. * If the agent already wrote its reply (as instructed in the prompt), th

Rough USD estimate for a transaction based on task payload.

* Execute a single field task via the HTTP API. * Tracks in-flight state and logs results.

Get the conversation thread for a message (by subject thread + taskId).

* Optional: Return financial metrics for this service. * Only implement if the service has financial data to display * (balances, revenue, spend

* Calculate retry delay with exponential backoff. * delay = retryDelayMinutes * 2^(attempt-1), capped at MAX_RETRY_DELAY_MINUTES

* Handle a failed task execution. * Pushes to the persistent retry queue with exponential backoff instead of setTimeout.

* Post-completion side effects: mark task done, post inbox message, log activity. * Each step is wrapped in its own try/catch — if one fails, others

* Lightweight, read-only connection test. * Makes a real API call to verify credentials are valid and the service is reachable. * Must not produ

* Check if a PID is still alive.

* Check if a PID is still alive.

* Check if a process is still running by sending signal 0.

* Kill a process tree. Uses tree-kill if available, falls back to process.kill.

* Kill a process tree. Uses tree-kill if available, falls back to process.kill.

* Kill a process tree.