donor-proposal-scoring

Warn

Audited by Gen Agent Trust Hub on Aug 5, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation in references/word-export.md contains Python code that performs dynamic path manipulation using sys.path.insert to load modules from a hardcoded local Windows directory.\n- [REMOTE_CODE_EXECUTION]: The skill automates the creation and execution of local Python scripts (e.g., gen_<slug>_docx.py) to handle Word document exports, representing a dynamic execution pattern.\n- [COMMAND_EXECUTION]: The instructions require the agent to run shell commands for managing git repositories and executing generated Python scripts.\n- [EXTERNAL_DOWNLOADS]: The skill depends on a non-standard local custom library (ane_package) stored at absolute system paths, which creates a dependency on an unverified and uncontrolled local environment.\n- [DATA_EXFILTRATION]: Multiple files contain hardcoded absolute Windows file paths (e.g., C:\\Users\\AGasser\\OneDrive\\...) that expose sensitive system information, including the local username and internal project directory structures.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection.\n
  • Ingestion points: The skill reads and processes external donor call documents and grant proposals (SKILL.md Step 1 and 2).\n
  • Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the ingested proposal text.\n
  • Capability inventory: The skill has capabilities to write files, perform git commits, and execute local Python scripts.\n
  • Sanitization: There is no evidence of sanitization or escaping of the external content before it is processed or used in script generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 5, 2026, 07:54 PM
Security Audit — agent-trust-hub — donor-proposal-scoring