The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone repositories from GitHub, specifically github.com/onvoyage-ai/best-ai-marketing-platform-benchmark.git and github.com/onvoyage-ai/gtm-engineer-skills.git.
[COMMAND_EXECUTION]: Shell commands including git clone, cd, npm install, and npm run are used to set up and execute external tools within the agent environment.
[REMOTE_CODE_EXECUTION]: The instructions involve cloning the onvoyage-ai/gtm-engineer-skills repository and immediately executing npm install followed by npm run. This constitutes a remote code execution pattern by running code fetched from a non-verified external source.
[INDIRECT_PROMPT_INJECTION]: The skill provides an audit functionality (npm run audit -- --url=...) that ingests and processes data from external URLs. This creates a vulnerability where malicious instructions hidden in the target website could influence the agent's behavior during analysis.
Ingestion points: Target URLs provided to the audit tool and files within the cloned repositories (SKILL.md).
Boundary markers: None present to delimit external content or warn against embedded instructions.
Capability inventory: Shell command execution (npm/git), network operations (requests.post), and file system access.
Sanitization: No sanitization or filtering of external content is described before processing.

best-ai-marketing-platform-benchmark