harness-sync
Pass
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by reading untrusted data from project manifest files and incorporating it into the agent's core instruction set.
- Ingestion points: In Step 3 of SKILL.md, the skill gathers repo facts and commands from files like package.json, Makefile, and CI configurations.
- Boundary markers: The skill uses
<!-- harness:begin -->and<!-- harness:end -->delimiters to scope its modifications, which helps identify managed content but does not prevent the inclusion of malicious instructions within that scope. - Capability inventory: The skill has the capability to write to primary agent instruction files (e.g., CLAUDE.md, AGENTS.md) and create project-level directories and documentation.
- Sanitization: There is no evidence of sanitization or validation of the strings extracted from manifests before they are placed into the generated instructions.
Audit Metadata