agent-architecture

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEOBFUSCATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [OBFUSCATION]: The skill consistently uses a non-standard character substitution where the '$' character replaces the '/' character in URLs, paths, and MIME types within documentation and configuration templates.
  • Evidence: 'https:/$api.example.com$v1', 'application$json', '$auth$login', and 'path: $health'.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests design data from a previous architectural phase via platform memory search, presenting a surface for untrusted instructions to influence the agent's behavior.
  • Ingestion points: The 'pre' hook in the frontmatter executes 'memory_search "pseudo_complete"'.
  • Boundary markers: No delimiters or specific instructions to ignore embedded commands are present for the ingested memory content.
  • Capability inventory: The skill can execute shell commands (via hooks), write to platform memory ('memory_store'), and generate structured system architecture documentation.
  • Sanitization: No sanitization or validation of the retrieved memory data is implemented.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands and pipes within its lifecycle hooks to manage state and search memory.
  • Evidence: 'memory_search "pseudo_complete" | tail -1' and 'memory_store "arch_complete_$(date +%s)"'.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 11:38 AM
Security Audit — agent-trust-hub — agent-architecture