main-stack-agents-flow
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill functions as an orchestrator that ingests potentially untrusted data from external sources and uses it to drive automated workflows.
- Ingestion points: User-provided feature requests and bug reports are ingested in the DESIGN phase (Step 1) to determine the architectural plan. Reports and diffs from implementer agents are ingested during the BUILD (Step 2) and VERIFY (Step 3) phases.
- Boundary markers: The instructions reference 'contracts' for task boundaries and 'scoped briefs' for punch-lists, but they do not explicitly command the AI to ignore or sanitize instructions that might be embedded within the project files or task reports.
- Capability inventory: The skill orchestrator can dispatch multiple domain-specific agents (Designer, Implementer, Verifier), write to a persistent 'Progress ledger' file on disk, read project architecture documentation, and manage transient state via a 'delete_memory' tool.
- Sanitization: No explicit sanitization or filtering of input data is defined. The skill relies on a 'Verifier' agent and human-in-the-loop approval to mitigate risks, which may not prevent sophisticated prompt injection attacks embedded in the processed data.
Audit Metadata