Quality Gate Checker

Problem

In agent workflows, marking a task as done does not guarantee delivery. A group-chat response may only be a confirmation phrase, or the output may lack code, links, and acceptable artifacts. Quality Gate Checker validates worker-submitted tasks and rejects nonconforming results for rework.

How It Works

The skill runs on a cron schedule (recommended every 15 minutes) or manual trigger, then reads completed tasks from the Dashboard API. It applies three checks:
- Completeness: very short replies, confirmation-only messages, or completed tasks without a reported output fail the gate.
- Timeliness: long completion time with minimal output triggers a warning; tasks over the maximum delay are logged and passed with a delay note.
- Consistency: tasks requiring push to CNB, articles, scripts, or Skills are checked for links, code blocks, or file paths; missing evidence can cause rejection or warning.
Rejected tasks are resent to the original worker’s inbox with a [Quality Reject] prefix, publicly noted in chat, and escalated to a VP after two consecutive rejections.

Boundaries

It is useful for rule-based acceptance of text, code links, articles, or Skill deliverables, but it does not verify code correctness, security, or business outcomes. Configure thresholds such as MIN_OUTPUT_LENGTH, REJECT_KEYWORDS, and MAX_ACCEPTABLE_DELAY to match team delivery patterns and avoid misjudging long discussions, cross-timezone work, or multi-person handoffs.