The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions for the agent to register recurring tasks using system and platform utilities such as cron, set_scheduled_task, and create-shortcut. This is used to facilitate a "weekly review" cycle where the agent autonomously processes accumulated observations.
[INDIRECT_PROMPT_INJECTION]: The skill operates by analyzing untrusted data sources including active session history, handoff documents, and user corrections to derive new rules or modify existing skill instructions. Ingestion points: Monitors real-time conversation history and reads a shared log.md file containing past observations. Boundary markers: Employs a structured logging format and mandates a "staging" step where updates are written to a temporary folder for user review rather than being applied directly. Capability inventory: The skill uses file write/read operations, shell command execution, and task scheduling capabilities to manage its library. Sanitization: Includes a multi-layered "Confidentiality layers" protocol designed to strip project-specific or sensitive details before generating reusable principles or skills.
[COMMAND_EXECUTION]: The skill utilizes shell commands including grep, sed, sort, tail, chmod, cp, and zip. These are used to manage its internal logs, handle concurrent file writes by multiple agent instances, and package multi-file skill updates for the user.

task-observer