After Deploy #449, the plan's upkeep asked flagon-io's provider (Anthropic, Auto) for #152, #94 and #59 at 04:30 and kept no suggestion for any of them, with nothing in the log: an answer that did not read as the four lines, and an empty one, were both marked as asked and never asked again. Now a JSON object reads too (in a code fence or not, duplicate or duplicate_of), keys with a heading mark or quotes read, an answer that still does not read is logged (clipped), and no answer at all is logged; suggestions and standings may use the most a brief may ask for (200 tokens, up from 120). Migration 0058_plan_ask_again asks again, once, for open work items in Triage that were asked and kept no suggestion.
Checked: cargo test -p g1t-work --lib (302, new cases: JSON answers, fenced and bare, and keys with marks); cargo check -p g1t-work --target wasm32-unknown-unknown. G1t-Session: ses_01m5bgsckce7gt943v64h219yy