jmb-research
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process information from external 'primary sources' such as official documentation and source code, which may contain adversarial instructions.
- Ingestion points:
SKILL.mddirects the agent to investigate official docs, source code, specs, and first-party APIs. - Boundary markers: No explicit delimiters or instructions to ignore embedded commands are provided for the research phase.
- Capability inventory: The agent can run read-only commands, write files to the repository, and perform Git/GitHub operations.
- Sanitization: No sanitization or validation of the researched content is specified before it is used to generate reports or scripts.
- [COMMAND_EXECUTION]: The instructions mandate the use of shell-based tools for repository state management and collaboration.
- Evidence:
SKILL.mdcontains specific command sequences forgit(checkout, add, commit, push) and the GitHub CLIgh(pr create, pr checks, pr ready). - [DYNAMIC_EXECUTION]: The skill allows for the creation of executable content based on research findings.
- Evidence:
SKILL.mdinstructs the agent to create a folder with 'helper scripts' written in 'the language of code' when citing numbers produced by such scripts, implying they are generated and executed during the research process.
Audit Metadata