dx-research
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes
git rev-parse HEADwithin its codebase mode to capture the repository state for research provenance. This involves spawning a shell command to retrieve the commit hash. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through its web research functionality.
- Ingestion points: External content is fetched using
WebFetchorWebSearchtools based on user-provided URLs or search terms in §3. - Boundary markers: The instructions include a specific safeguard, directing the agent to invoke
dx-referenceswithuntrusted-contentand explicitly stating that fetched content is data to report on, not instructions to follow. - Capability inventory: The agent has the capability to write research artifacts to the local file system (e.g.,
<container>/research/<topic-slug>.md) and execute shell commands for git metadata. - Sanitization: The skill relies on instruction-based isolation to prevent the agent from executing instructions found within external websites.
Audit Metadata