recommend-evolution

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection due to its design of persisting observation data to agent memory files.
  • Ingestion points: Untrusted data enters the agent context via the evidence and summary input fields defined in SKILL.md and processed by scripts/main.cjs.
  • Boundary markers: Recorded evidence is appended to .claude/context/runtime/evolution-requests.jsonl and .claude/context/memory/learnings.md. The implementation lacks explicit delimiters or instructional warnings to prevent the agent from interpreting instructions that may be embedded in this stored data when it is retrieved in future sessions.
  • Capability inventory: The skill is configured with Read, Write, Edit, and Skill tools, enabling it to modify local files and invoke other capability-rich skills such as artifact-integrator or skill-updater.
  • Sanitization: No sanitization or escaping is performed on the input strings before they are persisted to the filesystem, allowing potential prompt injection payloads to be stored in the agent's long-term memory.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 01:44 AM