The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the clawdhub package globally via npm. While the registry (NPM) is a well-known service, the package itself is provided by the skill author and is not from a pre-verified trusted organization.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of additional agent instructions and code from a remote registry (https://clawdhub.com). This enables the dynamic extension of the agent's capabilities with content from a third-party source.
[COMMAND_EXECUTION]: The skill uses various shell commands through the clawdhub CLI (search, install, update, publish, etc.) to interact with the host system and the remote registry.
[DATA_EXFILTRATION]: The clawdhub publish command allows the agent to upload local directories to the remote registry at https://clawdhub.com. This functionality, combined with network access to a non-whitelisted domain, could be used to exfiltrate sensitive data if the agent is directed to publish paths containing private information.
[INDIRECT_PROMPT_INJECTION]: The skill's ability to fetch and update instructions from an external source creates a surface for indirect prompt injection, where malicious skills hosted on the registry could influence agent behavior.
Ingestion points: Commands like clawdhub install and clawdhub update ingest external skill content into the local environment.
Boundary markers: No delimiters or safety instructions are defined to separate the tool output or the external skill content from the agent's primary instructions.
Capability inventory: The skill has capabilities for shell execution, file system modification, and network communication.
Sanitization: There are no explicit content verification or sanitization steps provided in the skill documentation.

clawdhub