The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and install executable binaries from untrusted sources using go install for the epa-echo-pp-cli and epa-echo-pp-mcp tools from github.com/mvanhorn/printing-press-library. Additionally, it uses npx to execute code from the @mvanhorn/printing-press-library package.
[EXTERNAL_DOWNLOADS]: The skill depends on external packages and source code fetched from the NPM registry and GitHub. These resources are not from established trusted organizations or well-known services, increasing the risk of supply chain attacks.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> feature that enables the transmission of command output to any external URL. Furthermore, the "Automatic learning" system journals user queries, command retries, and session data to a local teach.log and SQLite database. While the skill includes instructions to strip PII, this mechanism provides a persistent store of user interactions that could be accessed or exfiltrated.
[COMMAND_EXECUTION]: The skill's primary function is the execution of shell commands. It includes a which command that resolves natural-language queries into executable subcommands and utilizes shell backgrounding (&) for its teach and amend functions.
[PROMPT_INJECTION]: The skill processes data from the EPA ECHO API to generate compliance dossiers and comparative reports. This ingestion of external, potentially attacker-influenced data creates a surface for indirect prompt injection, especially as the skill lacks explicit sanitization or boundary markers for the processed content.

pp-epa-echo