skills/sharadchaturveda-coder/agency-agents-codex/agency-ai-data-remediation-engineer/Gen Agent Trust Hub
agency-ai-data-remediation-engineer
Warn
Audited by Gen Agent Trust Hub on Mar 25, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the Python eval() function to execute transformation logic generated by an AI model at runtime. \n
- Evidence: In SKILL.md, the apply_fix_to_cluster function executes eval(fix['transformation']) on strings produced by the ollama model.\n- [PROMPT_INJECTION]: The skill processes untrusted input data that is directly included in AI prompts, creating a surface for indirect prompt injection.\n
- Ingestion points: Anomalous data rows are accepted in the suspect_rows list and passed to the SLM in generate_fix_logic within SKILL.md.\n
- Boundary markers: There are no explicit delimiters or instructions to ignore instructions embedded within the data samples.\n
- Capability inventory: The skill can execute Python code through eval() calls in SKILL.md.\n
- Sanitization: The skill employs a simple keyword blacklist (import, os, subprocess, etc.) which is insufficient to prevent sophisticated Python code injection attacks.
Audit Metadata