iros-registry-automation

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions in instruction.md direct the agent to clone an external, third-party GitHub repository located at https://github.com/challengekim/iros-registry-automation.git to obtain the necessary automation scripts.
  • [REMOTE_CODE_EXECUTION]: The workflow involves executing Python scripts (such as iros_cart.py, iros_download.py, and iros_wizard.py) obtained from the external repository. The skill implements a security control by pinning the repository to a specific reviewed commit hash (7c6924b2ff88d693a12556659188cb91041e5097) defined in scripts/upstream.pin.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process sensitive legal documents (PDFs) which could serve as a vector for indirect injection.
  • Ingestion points: Document files located in the user-defined $workdir/downloads/ directory.
  • Boundary markers: The instruction.md and the summary script explicitly enforce a "facts only" policy, instructing the agent to extract data without providing legal interpretations or safety judgments.
  • Capability inventory: The skill utilizes file system read access and utilizes Python libraries (pypdf, pdfplumber) for text extraction.
  • Sanitization: The scripts/iros_pdf_summary.py script includes a mask_pii function that utilizes regular expressions to mask Resident Registration Numbers (RRNs) within the extracted text to prevent data exposure in logs or summaries.
  • [DYNAMIC_EXECUTION]: The script scripts/iros_pdf_summary.py uses dynamic imports to load either pdfplumber or pypdf based on their availability in the environment. Additionally, the test suite tests/test_iros_pdf_summary.py uses importlib.util to dynamically load the summary script for unit testing. The skill also utilizes the vendor's own CLI tool @nomadamas/k-skill to execute instructions and scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 07:49 AM