self-reflection

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted historical data from 'memories' and a 'ReasoningBank' to generate documentation improvements and trigger automated workflows.
  • Ingestion points: Untrusted data enters the context via memory_search queries and memory retrieval from the contextd MCP server.
  • Boundary markers: No specific delimiters or 'ignore embedded instructions' warnings are defined to separate retrieved memory content from the remediation instructions.
  • Capability inventory: The skill possesses significant capabilities including 'Apply changes', 'Create Issue/PR', and 'Run batch tests via subagents'.
  • Sanitization: There is no evidence of sanitization, validation, or escaping of the retrieved memory content before it is used to influence document updates or automated testing.
  • [REMOTE_CODE_EXECUTION]: The remediation flow involves 'Run batch tests via subagents' and 'Generate pressure scenarios'. If these tests or scenarios are dynamically constructed from untrusted memory data, they provide a mechanism for indirect code execution within the testing environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 06:39 PM