reproduce-and-fix-issues
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically Slack threads and tracker issues, which could contain malicious instructions designed to manipulate the agent's behavior.
- Ingestion points: Slack thread content and tracker issue data are read in sections 1, 2, 3, and 6 of SKILL.md.
- Boundary markers: The skill does not explicitly define delimiters or specific instructions to ignore embedded content within the ingested text.
- Capability inventory: The skill has capabilities for GitHub Pull Request creation (Section 14), Slack messaging (Sections 4, 9, 14), and UI automation via a control adapter (Section 5).
- Sanitization: No specific sanitization or filtering is described for the ingested content, although the skill includes extensive "Hard safety rules" (e.g., read-only workers, Slack write prohibitions for children) to limit the impact of potential injections.
- [DYNAMIC_EXECUTION]: The skill dynamically invokes other skills based on the
control.skill_nameparameter provided in the external Benny configuration. While this facilitates a pluggable architecture, it introduces a dependency on the integrity of the external configuration source. - [COMMAND_EXECUTION]: The skill relies on a control-adapter contract to perform complex system interactions, including starting application revisions, driving a real UI through clicks and keystrokes, and capturing media evidence, all of which involve executing system-level commands and manipulating the local environment.
Audit Metadata