auto-collaboration-dev-pipeline
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The sub-agent instruction templates in the
references/directory (e.g.,architect.md,backend-developer.md,requirements-analyst.md) use strict formatting constraints such as "Do not ask questions" and "Do not add any preface or epilogue." These are functional instructions intended to ensure structured output for the automated pipeline rather than malicious attempts to bypass safety filters. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user requirements and passes them through a chain of autonomous sub-agents, creating a surface for indirect prompt injection.
- Ingestion points: User input is ingested at the start of the pipeline (S1: Requirements Analysis) in
SKILL.md. - Boundary markers: The prompt templates use curly brace placeholders within markdown code blocks to delimit user input, providing basic structure but not high-assurance isolation.
- Capability inventory: The pipeline has the capability to write multiple source code and documentation files to the local workspace and includes instructions for the main agent to perform code integration and bug fixes.
- Sanitization: No explicit sanitization or filtering of user-provided requirement text is performed before it is interpolated into sub-agent prompts.
- [CODE_REVIEW]: The pipeline incorporates a specific 'Code Reviewer' role (
references/code-reviewer.md) designed to scan for security vulnerabilities like XSS, CSRF, and hardcoded credentials in the generated code, which serves as a significant defensive measure.
Audit Metadata