token-bounded-summaries
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is purely instructional and aimed at improving agent efficiency. No malicious patterns, exfiltration, or credential risks were detected.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external code diffs, which constitutes an ingestion surface for indirect instructions.
- Ingestion points: The
diff_contentvariable in the Python example (SKILL.md). - Boundary markers: The protocol uses a strict structural formula and a 60-token limit to constrain model output, though it lacks explicit input delimiters.
- Capability inventory: No dangerous tools or subprocess capabilities are associated with the summarization task.
- Sanitization: External input is interpolated into the prompt without explicit escaping.
Audit Metadata