The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted manifest files.\n
Ingestion points: The validate_wasm_bundle.py script reads a JSON manifest file provided via the --manifest argument.\n
Boundary markers: The instructions do not define clear delimiters or "ignore" instructions for the data contained within the manifest or the files it references.\n
Capability inventory: The script performs file read operations (path.read_bytes() and path.read_text()) on any path specified in the manifest, as seen in scripts/validate_wasm_bundle.py.\n
Sanitization: No path sanitization is performed; the resolve_path function allows absolute paths, enabling the script to read any file accessible to the agent process.

wasm-integration