tech-hub-platform
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The agent coordinates sub-agents by interpolating project context and task instructions into spawn prompts, which creates a potential surface for instructions embedded in project files to influence sub-agent behavior.
- Ingestion points: Data is ingested from the agent's initial spawn prompt (Step 1 in
AGENTS.md) and by reading specialist documentation files located at.claude/skill-docs/*.md(Step 2 inAGENTS.md). - Boundary markers: The agent prompt template (Step 4 in
AGENTS.md) uses labeled sections but lacks hard delimiters (such as XML tags or triple-dashes) or 'ignore' instructions to separate variable content from the sub-agent's core instructions. - Capability inventory: The skill can spawn autonomous sub-agents with the
Agent()tool, read project files, and coordinate tasks involving cloud infrastructure provisioning, secret management, and CI/CD configuration. - Sanitization: The instructions do not specify any sanitization, validation, or escaping of the project-specific metadata or task descriptions before they are used to prompt specialists.
Audit Metadata