plan_title: Multi-Model Analysis Example
task_title: Compare error-handling strategies across the runtime
task_description: |
Analyze how the Rhei runtime distinguishes recoverable from unrecoverable
failures during `rhei run`: where errors abort the orchestrator, where they
are surfaced as warnings, and where work is retried or skipped. Identify the
three weakest points in the current approach and, for each, name the file and
the concrete failure mode. Stay scoped to the runtime crates; do not propose
a redesign.
analysis_output_dir: runtime/analyses
agents:
- id: claude
label: Claude analysis note
selector: claude-code[yolo]:anthropic:claude-opus-4-7
- id: gemini
label: Gemini analysis note
selector: gemini[yolo]:google:gemini-3.1-pro-preview
- id: codex
label: Codex analysis note
selector: codex[yolo]:openai:gpt-5-codex
summary_agent: claude-code[yolo]:anthropic:claude-opus-4-7
final_document_path: runtime/final-analysis.md