oma-pdf
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill references a remote installation script for the
uvtool (https://astral.sh/uv/install.sh). This script is provided by Astral, a well-known technology provider, and is the official method for installing the package manager. As this involves a trusted source, the reference is considered safe practice. - [COMMAND_EXECUTION]: The skill instructions call for the execution of command-line tools using
uvx, specificallyopendataloader-pdffor document parsing andmdformatfor Markdown cleanup. These are standard operations for the skill's stated purpose of document conversion. - [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and parsing untrusted PDF files, which represents an indirect prompt injection attack surface. Malicious content within a processed PDF could potentially attempt to influence the agent's behavior.
- Ingestion points: The agent reads PDF files from the user-provided
input_pathas described inSKILL.mdandresources/execution-protocol.md. - Boundary markers: The instructions do not specify the use of delimiters or boundary markers to isolate the extracted PDF text from the agent's system prompt.
- Capability inventory: The skill allows the agent to write files to the local filesystem (
WRITE) and execute subprocesses viauvx(CALL_TOOL). - Sanitization: While the skill supports a
--sanitizeflag for PII masking, there is no evidence of sanitization designed to strip or neutralize potential prompt injection attacks from the document content.
Audit Metadata