Post-execution synthesis: combine executor outputs, call LLM with task_post_exec.md, parse into post_summary and out_doc; validate links. Retry on parse/validation issues: refill with previous output and issues via header_raw (same pattern as refinement).