clean-user-facing-text

Pass

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to run included Python scripts (e.g., clean_text.py, inspect_text.py) to process user-provided files.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted prose from external files and performs a rewrite operation, which is a significant surface for indirect prompt injection. Malicious instructions embedded in the text could attempt to hijack the agent's behavior during the rewrite phase.\n
  • Ingestion points: read_text_input in scripts/common.py reads user files into the context for processing.\n
  • Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands during the LLM-driven rewrite step.\n
  • Capability inventory: The agent has the ability to read and write files and execute shell commands to run its internal utility scripts.\n
  • Sanitization: While the skill performs Unicode filtering to remove invisible artifacts, it does not sanitize the semantic content of the prose for injection payloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 31, 2026, 08:17 PM
Security Audit — agent-trust-hub — clean-user-facing-text