The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill facilitates the installation of the epa-echo-pp-cli and epa-echo-pp-mcp tools from the author's GitHub and NPM repositories (@mvanhorn/printing-press-library). While these are vendor-owned resources, they involve downloading and executing remote code during the setup process.
[COMMAND_EXECUTION]: The skill frequently executes the epa-echo-pp-cli binary via bash to perform facility resolution, compliance evidence inspection, and data synchronization tasks.
[DATA_EXPOSURE_AND_EXFILTRATION]: Provides a --deliver webhook:<url> flag that enables routing command results to arbitrary external URLs. Additionally, a feedback mechanism is included that can send data to a remote endpoint if EPA_ECHO_FEEDBACK_ENDPOINT is configured in the environment.
[DYNAMIC_EXECUTION]: Features a 'Playbook' system where sequences of CLI commands are dynamically assembled and executed based on steps defined in JSON files. These playbooks can be automatically synthesized from the agent's session journals.
[INDIRECT_PROMPT_INJECTION]: The 'Learning Loop' (via recall and teach commands) creates a vulnerability surface by ingesting user queries and external API results to influence future agent behavior and automated playbooks.
Ingestion points: EPA ECHO API results and user-supplied natural language queries processed by the recall command in SKILL.md.
Boundary markers: The instructions recommend stripping personal identifiers (PII) before recording data, but no explicit delimiters are defined for the playback of generated command sequences.
Capability inventory: The environment includes full bash access, network connectivity via the CLI and webhooks, and the ability to write to local configuration and data directories (SKILL.md).
Sanitization: Relies on the agent following instructions to manually clean and validate data before invoking the teach or amend commands to update the local learning store.

pp-epa-echo