The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided regulatory documents (PDFs or text). This represents an indirect prompt injection surface. The skill mitigates this risk by instructing the agent to extract requirements into a structured JSON format and requiring explicit human confirmation of these requirements before performing any automated analysis. This process ensures that any instructions embedded within the processed documents are reviewed by a human before being acted upon by the agent.
[COMMAND_EXECUTION]: The skill utilizes several local scripts for data processing and analysis. The scripts/artifact_paths.py script correctly implements shell-escaping for file paths (escaping single quotes) when generating output meant for shell evaluation (eval), which prevents potential command injection if a user provides a maliciously crafted filename containing shell metacharacters.
[EXTERNAL_DOWNLOADS]: The skill interacts with the official Sumsub API (api.sumsub.com) to resolve configuration graphs. It implements a credential guard in scripts/sumsub_curl.sh that validates the SUMSUB_APP_TOKEN environment variable, ensuring it has a sandbox prefix (sbx:) and warning against the use of production credentials, which helps prevent accidental exposure of sensitive live keys.

sumsub-conformance