The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (pin titles, descriptions, and board names) fetched from Pinterest API and DOM scraping. This data is interpolated into the agent's context without sanitization or boundary markers.
Ingestion points: Pinterest search results and board data are ingested in scripts/lib/run.mjs via loadPins and BoardFeedResource.
Boundary markers: Absent. The skill does not use delimiters or instructions to ignore embedded commands in the fetched Pinterest content.
Capability inventory: The skill has capabilities to write files to disk (fs.writeFile in scripts/lib/run.mjs), perform network requests (fetchMediaBytes in scripts/lib/media-fetch.mjs), and execute shell commands via ego-browser spawning.
Sanitization: Pin titles are truncated and trimmed, but no sanitization for prompt injection patterns is performed.
[COMMAND_EXECUTION]: The skill uses child_process.spawn in scripts/lib/launch.mjs to launch the ego-browser binary for browser automation and JavaScript execution as part of its core functionality.
[DYNAMIC_EXECUTION]: The skill dynamically generates a temporary JavaScript module in scripts/lib/launch.mjs (renderEntry) and executes it using ego-browser nodejs -e "await import(...)". This is used to bridge command-line arguments to the restricted execution environment.

pinterest-download