The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to upload local PDF documents to a remote server located at http://39.104.68.74:8082 (and mentions http://192.168.41.78 in scripts). This involves sending user-provided content to an external endpoint that is not a well-known or trusted service.
[EXTERNAL_DOWNLOADS]: The skill requires the httpx Python package to be installed from a public registry to handle API communications.
[PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it processes schemas and extraction results generated by a remote AI engine. Maliciously crafted documents or a compromised API could return instructions disguised as data to influence the agent's behavior.
Ingestion points: Results and schemas consumed via Server-Sent Events (SSE) in scripts/client.py (notably in chat_schema, extract, and chat_correct methods).
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard malicious commands embedded within the extracted data.
Capability inventory: The skill can read local files and perform network requests to the specified API.
Sanitization: No evidence of sanitization or validation of the remote AI's output before it is integrated into the agent's context.
[COMMAND_EXECUTION]: The skill provides a CLI script (scripts/extract_flow.py) and expects the agent to execute shell commands to manage the document extraction workflow, including uploading files and retrieving results.

doc-extract-engine