[Phase 1] ReportGenerationAgent (purely generative, no tools)

[varun-polpakkara]

Summary

LangGraph agent called by Stage 5 to synthesise the two-audience Markdown reports. Purely generative — receives structured data, produces narrative. No tools (no file reads, no web access).

Interface

class ReportGenerationAgent:
    def __init__(self, llm: BaseChatModel, *, max_iterations: int = 10): ...
    def generate(self, scan_data: ScanReportData) -> ReportOutput: ...

ScanReportData: snapshot metadata + all components + all findings (with reachability + exploit details) + all triage history.

ReportOutput: engineering_report: str + executive_summary: str.

Constraints

• No tools — the agent only has access to the data it is passed
• If either output is empty string, ReportGenerationError is raised by Stage 5
• Must produce ISO 21434 and UNECE R155 framing in the executive summary

LLM provider

Same BaseChatModel instance injected by the CLI — configured via CYBERGUARD_LLM_PROVIDER env var.