bento-maintain
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process various external data sources from the target repository which could contain malicious instructions.
- Ingestion points: The skill reads repository-level instruction files (
AGENTS.md,CLAUDE.md), localized instruction overlays (.agents/bento/bento-maintain.md), and monitoring logs or traces (references/error-fix.md). - Boundary markers: No explicit instructions are provided for wrapping external data in delimiters or using boundary markers to isolate untrusted content from the agent's instructions.
- Capability inventory: The skill has the capability to modify files within its assigned scope and execute shell commands for verification, such as test runners (
python3 -m unittest). It also delegates tasks to other high-capability skills likebento-refactorandbento-investigate. - Sanitization: The instructions do not specify sanitization or validation of the content of the processed files before they are used to influence the agent's maintenance tasks.
Audit Metadata