The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the clawhub CLI for various management tasks such as searching, listing, and updating skills.
[EXTERNAL_DOWNLOADS]: The skill performs a global npm installation of the clawhub package and connects to https://clawhub.com for registry operations.
[DATA_EXFILTRATION]: A test case in SAFETY.md references a potential exfiltration path using pastebin.com and wget. Analysis confirms this is part of safety documentation describing behavior the agent must refuse, rather than an active instruction.
[PROMPT_INJECTION]: The skill processes skill metadata from an external registry, representing a surface for indirect injection.
Ingestion points: External data is ingested through the clawhub search and clawhub install commands in SKILL.md.
Boundary markers: The SKILL.md file contains a dedicated 'Security Guardrails' section with explicit constraints on agent behavior.
Capability inventory: The skill allows for global package installation and file system modification via CLI commands.
Sanitization: The skill relies on explicit agent instructions and user confirmation rather than automated sanitization of remote metadata.

clawhub-hardened