medication-adherence-message-gen
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data through command-line arguments in
scripts/main.py, which are then interpolated into message templates. - Ingestion points: Command-line arguments (
--name,--medication,--dosage,--time) inscripts/main.py. - Boundary markers: The script does not use specific delimiters or instructions to prevent the agent from obeying instructions that might be embedded in patient or medication names.
- Capability inventory: Analysis of
scripts/main.pyconfirms no network access, no file system writes, and no command execution capabilities are present. - Sanitization: No input sanitization or validation is applied to the user-supplied strings before they are formatted into the final output message.
- [METADATA_POISONING]: The skill package includes a file
medication-adherence-message-gen_audit_result_v2.jsonwhich provides a simulated security audit. - Evidence: This file contains pre-canned "PASS" results for security gates and stability, which is a self-referential attempt to influence the agent's or analyzer's safety evaluation.
- Evidence: The
SKILL.mdfrontmatter identifies the author asAIPOCH, while the script comments and markdown footer creditOpenClaw, representing inconsistent metadata that could mislead attribution.
Audit Metadata