The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell command execution to interact with Google APIs when a proxy is detected.
Evidence: In scripts/providers/google.ts, the postGoogleJsonViaCurl function uses execSync to run a curl command. The command string is constructed using environment variables (HTTPS_PROXY, ALL_PROXY, etc.) without sanitization. If these variables contain shell metacharacters or quotes, it could lead to arbitrary command execution.
[EXTERNAL_DOWNLOADS]: The skill communicates with several external AI service providers to generate and retrieve image data.
Evidence: The scripts make network requests to official endpoints for Google Gemini, OpenAI, Alibaba DashScope, and Replicate. These are well-known technology services used for their intended purpose in this skill.
[CREDENTIALS_UNSAFE]: The skill is configured to load API credentials from local environment files.
Evidence: scripts/main.ts includes a loadEnv function that reads .env files from the current project directory (.beaver-skill/.env) and the user's home directory (~/.beaver-skill/.env) to populate process environment variables with API keys.
[PROMPT_INJECTION]: The skill processes external data that could contain indirect prompt injections.
Evidence: The skill ingests untrusted data via --promptfiles and --ref parameters. While it stringifies this data for JSON API payloads, it lacks explicit boundary markers or sanitization to prevent the generated image models from following instructions embedded in the provided text or image metadata.

beaver-image-gen