The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install a custom CLI tool directly from a remote GitHub repository using uv tool install "git+https://github.com/lirrensi/agent-sommelier". This executes code from an external source during the installation process.
[EXTERNAL_DOWNLOADS]: The skill documentation encourages the use of npx skills add <source> to download and install additional skills. This utility allows for fetching content from arbitrary remote sources, which may include unverified third-party repositories.
[REMOTE_CODE_EXECUTION]: The skill-store sync command described in references/importing.md automatically extracts .skill files (ZIP archives) placed in the store directory. Automatic extraction of archives from potentially untrusted sources can be a vector for directory traversal attacks (Zip Slip) if the underlying CLI tool does not perform adequate path validation.
[COMMAND_EXECUTION]: The "YOLO" autonomy mode described in references/organizing.md instructs the agent to automatically execute multiple shell commands (e.g., skill-store groups create, skill-store groups add) to reorganize the local filesystem without explicit user confirmation. This reduces oversight and could lead to unintended modifications of the local environment.
[PROMPT_INJECTION]: The skill's primary function is to load and preview external SKILL.md files into the active agent context. This creates a significant surface for indirect prompt injection (Category 8), where a malicious skill file could contain hidden instructions that override agent behavior once read.
Ingestion points: External skills are added via npx skills or manual file copies to ~/.skill-store/skills/.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are documented for the load or preview commands.
Capability inventory: The skill uses a CLI tool that can modify the filesystem and interacts with npx for network-based downloads.
Sanitization: There is no evidence of content sanitization or validation performed on loaded skill files before they are processed by the agent.

skill-store