The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill frequently executes npx goose-aeo@latest, which downloads and runs a package from the public NPM registry at runtime. Because it uses the @latest tag, the executed code can change without user notification or version pinning, creating a potential supply chain risk from a non-trusted source.
[INDIRECT_PROMPT_INJECTION]: The skill features a large attack surface for indirect prompt injection by scraping external websites (during the audit phase) and processing unstructured responses from multiple AI search engines (during the analyze phase).
Ingestion points: Website content scraped by the tool and responses returned from Perplexity, OpenAI, Gemini, Grok, Claude, and DeepSeek.
Boundary markers: None identified in the skill instructions to separate external data from system instructions or to warn the agent about potentially malicious content.
Capability inventory: The agent can execute shell commands via npx and write to local files like .env and .goose-aeo.yml.
Sanitization: No evidence of sanitization, validation, or filtering of external content before processing or generating recommendations.
[DYNAMIC_EXECUTION]: The skill uses node -e to execute a dynamically generated JavaScript string for checking environment variables. While used here for discovery, this pattern involves executing code assembled within the prompt.
[COMMAND_EXECUTION]: The skill relies heavily on shell commands including npx, cat, and echo to manage configuration, execute the primary tool, and check the environment.

aeo