The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch compressed archives (.tar.gz and .zip) from an external registry.
Evidence: Downloads are performed from https://registry.clawhub.ai/v1/skills/SLUG/download using curl -fsSL.
[COMMAND_EXECUTION]: The skill executes shell commands to manipulate the local filesystem and extract remote archives.
Evidence: Uses mkdir -p, tar xzf, unzip -o, and rm to manage skill installation in the skills/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and processing untrusted data from a remote web service, creating an attack surface for indirect prompt injection.
Ingestion points: Fetches skill listings and descriptions from https://clawhub.ai via web_fetch, and downloads full skill content from the registry.
Boundary markers: The skill relies on natural language instructions (the "Safety rule") to identify malicious patterns rather than using technical delimiters or sandbox isolation.
Capability inventory: The skill allows for command execution (exec), file writing (write), and network access (web_fetch), providing a significant impact path for any injected instructions found in downloaded skills.
Sanitization: No programmatic validation of the downloaded archive's contents is performed beyond the agent's manual keyword check and user confirmation step.

clawhub