reflect

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves reading and auditing session transcripts. These transcripts are untrusted data sources as they contain arbitrary user input. An attacker could inject instructions into the conversation history intended to deceive the review panel into proposing malicious edits to the agent's skill library.
  • Ingestion points: The transcript file path is located and read by three independent reviewers in SKILL.md (Workflow Step 1).
  • Boundary markers: Absent. The instructions do not define specific delimiters or guidelines to help the reviewers distinguish between legitimate session history and adversarial instructions embedded within that history.
  • Capability inventory: The skill is designed to propose concrete edits to the file system (specifically in skills/<name>/ or .claude/skills/<name>/) and utilizes the skillify tool for executing these changes after user approval.
  • Sanitization: Absent. The skill relies on the logic of the 'Judgment' reviewer and the final human user to detect and reject inappropriate or malicious modification proposals.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 07:21 PM