nexus
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to automatically extract and execute shell commands (test, lint, build, smoke-check) from a repository's instructions, CI configurations (e.g., GitHub Actions, GitLab CI), and manifests. This is a core function but allows for the execution of arbitrary code if the repository being processed is malicious.
- [INDIRECT_PROMPT_INJECTION]: The nexus agent processes untrusted data from multiple sources including repository documentation, CI scripts, and logs from sub-sessions. This creates a surface where a malicious repository or a compromised sub-session could inject instructions into the nexus's decision-making loop.
- Ingestion points: Extracts build/test logic from
env.md(source: project instructions/CI), reads sub-session logs insessions/*.md, and parses requirements fromincrement.json(source: design files). - Boundary markers: The skill implements a 'decisions register' and a rule to 'judge the record, not the chat' to separate claims from evidence, though it lacks explicit sanitization for extracted shell commands.
- Capability inventory: Spawns sub-sessions, executes git operations (worktrees, branch management), and runs arbitrary shell commands defined in the environment configuration.
- Sanitization: No specific sanitization or validation of the commands extracted from repository manifests is described, relying instead on the user to review the 'digest' of actions.
Audit Metadata