The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is designed to dynamically locate, load, and execute instructions from other SKILL.md files stored on the local filesystem. This capability allows the agent to switch between different workflows by reading content from computed file paths.
[COMMAND_EXECUTION]: The agent is instructed to execute a bundled Python script (scripts/discover_skills.py) to perform local discovery of available skills. While this script primarily performs filesystem scanning and integrity verification, it constitutes automated command execution.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where it ingests and executes instructions from external files (SKILL.md of discovered skills). 1. Ingestion points: SKILL.md files of discovered skills in local directories like ~/.agents/skills, ~/.codex/skills, and ~/.claude/skills. 2. Boundary markers: The instructions do not define explicit delimiters or warnings to separate the vsc instructions from the target skill's instructions. 3. Capability inventory: The skill allows the agent to execute any tools, scripts, or network operations defined within the target skill's SKILL.md file. 4. Sanitization: The skill uses a catalog.json file to verify the SHA256 checksums of discovered skills, though it permits reading the files even when the checksum does not match.

vsc