optimize-agent-docs
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION] (HIGH): Indirect Prompt Injection surface (Category 8).
- Ingestion points: The skill reads all markdown files within dotfile directories (
.claude/,.cursor/, etc.) and root instruction files (CLAUDE.md,INSTRUCTIONS.md). - Boundary markers: There are no boundary markers or instructions to ignore embedded commands within the processed data.
- Capability inventory: The skill generates a
KNOWLEDGE.mdfile designed to be the agent's 'always-loaded' context. If the source documentation contains malicious instructions (e.g., hidden in HTML comments or prose), the agent is instructed to extract them as 'Critical Rules' or 'Procedures'. - Sanitization: None. The skill assumes all content in the scanned documentation is authoritative and safe, creating a persistent 'instruction poisoning' vector.
- [COMMAND_EXECUTION] (LOW): Local shell command usage.
- Evidence: The workflow utilizes
find,grep, andwcto perform file discovery and analysis. While these are standard utilities, they are executed via the agent's shell capability on the local filesystem.
Recommendations
- AI detected serious security threats
Audit Metadata