The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements an 'Automatic learning' system where the recall command retrieves previously stored playbooks and notes from a local SQLite database to guide the agent's behavior. This creates a surface for indirect prompt injection if malicious or unexpected instructions are saved to the store and later retrieved. Ingestion points: Data returned by the groq-pp-cli recall command, specifically the playbook and notes fields. Boundary markers: The skill includes a 'SAFETY' section explicitly warning that recall-surfaced text is untrusted. Capability inventory: Uses Read Bash to execute shell commands with user-provided queries and retrieved playbook data. Sanitization: The agent is instructed to use single-quoting and specific escaping ('\'') for all dynamic values.
[COMMAND_EXECUTION]: The skill relies on shell execution of the groq-pp-cli binary for all functions. Several commands (e.g., recall, teach, playbook amend) interpolate user-provided strings into shell arguments. While the skill provides detailed instructions on how to handle these inputs safely, improper implementation by the agent could lead to command injection.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the groq-pp-cli and groq-pp-mcp tools. It fetches the @mvanhorn/printing-press-library package from the NPM registry and downloads Go binaries from the github.com/mvanhorn/printing-press-library repository. These are identified as vendor resources associated with the author.
[DATA_EXFILTRATION]: The skill includes a --deliver webhook:<url> feature that allows routing command results to an external URL via HTTP POST. While a documented feature, it could be leveraged for exfiltration if the choice of URL is influenced by an attacker.

pp-groq