beans
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes several local commands including
git,python3,jq, and thebeansCLI tool to perform task management operations and data processing. Evidence:git log,git rev-parse,beans list,beans graphqlinbin/citum-bean. - [DYNAMIC_EXECUTION]: The
bin/citum-beanscript uses an embedded Python script (python3 -c) to parse git log output into a structured JSON format. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from git commit history and local task files which could contain untrusted instructions.
- Ingestion points: Git commit messages via
git logand task files in the.beans/directory (file:bin/citum-bean). - Boundary markers: No explicit delimiters or instructions are used to warn the model about potential instructions embedded in the task data or commit history.
- Capability inventory: The skill allows the agent to read, create, and update task files via the
beansCLI (file:SKILL.md). - Sanitization: Data is parsed and structured using Python and
jqbefore being presented to the agent context.
Audit Metadata