The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script scripts/stitch_full_page_capture.mjs ingests untrusted data from a manifest.json file which defines the capture parameters.
Ingestion points: The script reads pageUrl, fullPageImage, and sectionImages[].file from the user-provided manifest file.
Capability inventory: The script utilizes Playwright for automated web browsing and ffmpeg for writing image files to the disk.
Boundary markers: There are no delimiters or instructions to ignore potentially malicious content within the manifest data.
Sanitization: The script does not perform path sanitization or validation on the image output paths, which may allow an attacker to use directory traversal sequences (e.g., ../../) to overwrite sensitive files outside of the intended project directory.
[COMMAND_EXECUTION]: The utility executes system commands ffmpeg and sips using node:child_process. Although the implementation uses safe argument-passing patterns, the lack of validation on the file paths sourced from the manifest means the skill could be used to read from or write to unintended locations on the host system.
[EXTERNAL_DOWNLOADS]: The helper script dynamically loads the playwright library from the local environment's node_modules. This introduces a dependency on external code that must be installed by the user, which is a standard but noteworthy surface for supply chain risks.

stitched-full-page-capture