grace-setup-subagents
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious scripts, unauthorized data access, or network exfiltration patterns were detected. The skill consists entirely of instructional markdown and configuration defining a software development methodology.
- [INDIRECT_PROMPT_INJECTION]: The subagent roles defined in the skill ingest and process external project data such as code and logs, presenting a standard vulnerability surface managed by procedural constraints.
- Ingestion points: Roles in 'references/roles/' ingest module implementations, failure packets, and '.grace' metadata files (SKILL.md).
- Boundary markers: The instructions mandate specific delimiters like 'MODULE_CONTRACT', 'START_CONTRACT', and 'END_CONTRACT' to differentiate instructions from code data.
- Capability inventory: Subagents are restricted to reading/writing project files and generating text; no network tools, shell execution, or high-privilege operations are enabled.
- Sanitization: The methodology relies on structured artifact processing and manual review mindset instructions rather than explicit programmatic sanitization of ingested text.
Audit Metadata