aii-job-poke

Claims and executes one due scheduled job from a tenant's shared queue per run.

Updated Aug 21, 2026
One-click install
npx skills add https://github.com/maestroventures/aiintegrator --skill aii-job-poke-maestroventures
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: aii-job-poke
Source: https://github.com/maestroventures/aiintegrator/tree/main/plugins/aii-blueprint/skills/aii-job-poke
Command: npx skills add https://github.com/maestroventures/aiintegrator --skill aii-job-poke-maestroventures

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Scheduled jobs across multiple AI platforms need a single, identical executor that runs each job exactly once without duplicating work or silently failing when a queue is unreachable. This Skill provides that recurring executor: it checks queue reachability, reaps abandoned jobs, claims exactly one due job, runs it, and records a verifiable outcome beat. ## Core Features & Use Cases - Exactly-once job claiming: Uses the Blueprint connector's job_poke_* tools so the database decides which executor gets a job, even when several platforms poke simultaneously. - Abandoned-job recovery: Reaps expired leases from dead executors so interrupted work returns to the queue instead of being lost. - Traceable execution beats: Every run records the seat, routine trigger id, body reference (door=), and outcome, distinguishing a healthy idle run from a silent failure. - Use Case: A company signs into Claude and other AI platforms; each platform runs this identical hourly poke, and whichever wakes first claims the due job (e.g., a daily CRM sync), executes it, and beats ok with the exact body reference it ran. ## Quick Start Install this skill once per AI platform by creating an hourly schedule named "AII Poke · <platform> · <account email>" whose text invokes the aii-job-poke skill, with all listed Blueprint tools allowed.

Frequently Asked Questions about aii-job-poke

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run scheduled jobs across multiple AI platforms exactly once?

Install one identical recurring poke per platform that calls the queue's claim function. The database decides which executor gets each job, so concurrent pokes never duplicate work; executors that get nothing report it as a normal successful outcome.

How do I recover jobs abandoned by a dead executor?

Call the reap function with your tenant id before claiming. It frees expired leases from executors that died mid-run and returns the affected jobs to the queue, turning an executor's death into a recorded warning instead of lost work.

Why does an unattended scheduled run stop on its first tool call?

The run hit a tool it was not allowed to use and is waiting on a permission prompt nobody will answer. Allow every required tool at schedule creation, or set them to Always allow in the connector settings, then verify by reading the schedule back.

What happens when the job queue is unreachable on a seat?

The executor checks reachability first and stops with a plain statement that no queue is reachable. It never improvises a fallback store or runs a partial workflow, because a silent no-op is indistinguishable from a healthy idle run.

Can a cloud-run executor open job bodies stored as file paths?

No. A cloud session has no file bridge, so a path-based body reference must be refused with a died-in-gate beat naming the path. Only sql: body references, read from the tenant's own store, are runnable by any executor.