What problem does it solve? Setting up monitoring usually means writing scripts or waking a model on every poll, which is expensive and error-prone. This Skill turns a plain-language request like "tell me when a log line contains ERROR" into a deterministic rule tree that a watcher evaluates mechanically at every poll, at zero model cost. ## Core Features & Use Cases - Rule tree generation: Builds JSON rule trees from combinators (et, ou, non) and deterministic leaves such as contient, heure_entre, jour_semaine, est_down, apparu, and nouvelle_ligne, matched to the correct watcher_type (log, file, url, command). - Watcher lifecycle management: Guides creation, listing, pausing, and deletion of watchers via watcher_create, watcher_list, watcher_toggle, and watcher_delete, including duplicate detection and never-fire validation feedback. - Correlation and actions: Supports correlating multiple watchers into one diagnosis and choosing the right fire action (agent, notifier, commande, aucune) to avoid unnecessary model calls. - Use Case: A user asks to be pinged if release.log contains ERROR but not at night; the Skill produces a log watcher with an et tree combining contient and an heure_entre window, firing only during working hours. ## Quick Start Ask the agent to notify you when a specific file appears on your desktop, and it will design and create the corresponding watcher rule tree.
