What problem does it solve? Long-running autonomous research pipelines leave users in the dark about progress, failures, and blockers. This Skill turns workspace artifacts into human-readable progress briefings and delivers them through whatever notification channel the user has already configured, so researchers stay informed without polling logs. ## Core Features & Use Cases - Opt-in briefing generation: Reads task.md for a notification directive and silently does nothing unless the user opted in, guaranteeing zero side effects otherwise. - Channel-agnostic dispatch: Detects whatever notification service is configured (email account, webhook, chat integration) and sends through it without hard-coding any tool; always saves a timestamped copy to notification/ even when nothing is configured. - Delta-aware reporting: Compares against the previous briefing to report what changed — runs finished, verdicts flipped, GPU-hours consumed — and leads with blockers requiring human action. - Use Case: During a multi-hour mechanistic interpretability experiment, the pipeline calls this Skill hourly and on done/halted/approval-needed events, emailing the researcher a TL;DR update like "C1 verified PASS, 2 runs left" with a clear action item when blocked. ## Quick Start Ask the agent to send a progress notification for the current run, for example by invoking /notify with an event type such as progress or approval-needed.