reflect
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted session transcript data to generate suggestions, creating a surface where instructions embedded in the session history could influence the agent's output.
- Ingestion points:
SKILL.md(Step 1) specifies locating the workspace transcript or writing a digest of the session history. - Boundary markers: No specific delimiters or warnings to ignore instructions within the transcript are defined.
- Capability inventory: The skill proposes file modifications that are executed via the
author-skillstool (Step 5) and includes a subprocess-like sequence through its lens subagents. - Sanitization: No content filtering or validation is specified for the transcript data before it is processed by the analysis subagents.
Audit Metadata