self-review

Pass

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill enables the assistant to ingest untrusted data from historical conversations to guide behavioral changes.
  • Ingestion Points: The skill processes current and historical chat sessions in Step 1 and Step 2 of the 'Historical Mode' (SKILL.md).
  • Boundary Markers: Absent. The instructions lack delimiters or constraints to prevent the agent from being influenced by instructions found within the logs it is analyzing.
  • Capability Inventory: The agent has the capability to propose and implement (with user consent) modifications to sensitive control surfaces such as CLAUDE.md (global instructions), memory, and path-scoped rules (SKILL.md).
  • Sanitization: No sanitization or verification mechanism is described for the ingested conversation data.
Audit Metadata
Risk Level
SAFE
Analyzed
May 17, 2026, 10:58 AM
Security Audit — agent-trust-hub — self-review