bootstrap-agentic-system
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a sophisticated bootstrapping framework for repository-local agentic systems. It focuses on setting up structured workflows for planning, implementation, and testing.
- [SAFE]: Implements strong context isolation and boundary enforcement (as described in
contracts/context-boundaries.md) to separate maintenance records, project knowledge, and runtime agent instructions. The maintenance root is explicitly excluded from ordinary agent search and discovery. - [SAFE]: All sensitive operations, such as code implementation, file deletion, and issue tracker updates, are protected by mandatory human-in-the-loop approval gates. Agents are instructed to stop and wait for explicit approval before proceeding with writes.
- [SAFE]: The system uses a "normative rules inventory" approach to ground agents in provided project guidelines. This mechanism requires agents to extract and follow explicit rules from knowledge files, preventing them from being swayed by potentially non-compliant or malicious patterns in the existing codebase.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources like issue trackers and repository files, which represents an indirect prompt injection surface. This is mitigated by the structured gate-based workflow, mandatory user approvals, and the grounding of agent behavior in local policy files. The assessed severity for this category is low.
Audit Metadata