The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and install content from external GitHub repositories using Git and the raw.githubusercontent.com domain for metadata extraction.
[COMMAND_EXECUTION]: The core logic utilizes shell commands such as ln -s, cp -R, mkdir, rm, and git to manage the lifecycle of agent skills and commands on the host system.
[DYNAMIC_EXECUTION]: The skill orchestrates multiple internal shell and Python scripts (record.py, security.py, target.sh) to perform installation, updates, and safety audits.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata (descriptions and version numbers) from external Markdown files found in third-party repositories.
Ingestion points: record.py reads and parses SKILL.md and CHANGELOG.md from local paths and remote URLs.
Boundary markers: No explicit prompt boundaries are used; metadata is extracted via regular expressions.
Capability inventory: The skill possesses capabilities for file system manipulation (ln, cp, rm), network access (git, urllib), and subprocess execution.
Sanitization: The skill includes a dedicated security.py module that scans all installed content for malicious patterns (e.g., prompt injection, credential exfiltration, RCE) before finalization.

skill-manager