langchain-architecture
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external documentation and repository evidence via the 'sandbase_discover' and 'sandbase_inspect' tools defined in 'references/sandbase-api-map.md'. This data enters the agent context, creating a potential vector for indirect prompt injection. The skill mitigates this risk by requiring the agent to verify all implementations with focused tests and source comparisons (SKILL.md).
- Ingestion points: External documentation and repository evidence from SandBase API tools.
- Boundary markers: General operational boundaries are defined in the 'SandBase boundary' section, but specific prompt delimiters are not provided.
- Capability inventory: The agent can execute capabilities through 'sandbase_run' and perform local verification tests.
- Sanitization: No explicit sanitization of external content is mentioned.
- [COMMAND_EXECUTION]: As part of its quality gates and workflow, the skill requires the agent to execute focused tests, build checks, and static analysis on its implementation (SKILL.md). This involves running code locally to ensure work satisfies acceptance checks.
- [DATA_EXFILTRATION]: The skill sends data to the SandBase API, which belongs to the skill's author. It includes instructions to transmit only the minimum necessary data and mandates user confirmation before performing paid calls, which serves as a control against unintended data exposure and costs.
Audit Metadata