{"content":"fix: widen agent-game waitFor budgets so slow hosts do not flake the chain (IDBots ecb03b5fa)\n\nRun 44495819432's Windows job cleared every earlier hurdle (NSIS gate, the panel locator pin, the session log guard fixtures — the whole group-tasks / skill-assignment / cowork-proxy / proposals / bot-browser span) and then died in 'Run agent-game tests' on a load-timing flake, not a product bug: 'action writes are signed as the session agent bot (mover pays), not the owner' waited 15s for the peer seat.claimed → turn → LLM-stub → action-write async chain and timed out at 15.1s under full-chain CI load on the self-hosted win-sign runner. The same suite run standalone on that machine passes 5/5 in seconds, which isolates the failure to host load, not Windows semantics.\n\nThis is the known §3.3 third-check class (fixed-ms budgets asserting async chains). Changes, assertions untouched — only the budgets: agentGameSeatJoinRpc's waitFor default 8s → 20s, the action-write wait 15s → 25s, test-level timeouts 20s/25s/30s → 45s/60s/45s so a wait can always expire inside its own test budget; agentGameSeatClaim's retry-to-commit wait 15s → 25s and its test 20s → 45s — that one carries its own comment about an earlier Windows CI flake (run 36427382778) in the exact same persisted-budget window, so its margin was demonstrably thin.\n\nVerified: both files green in the fix worktree (5/5 and 3/3, install + compile:electron first). Deliberately NOT touched: the other six agent-game files passed under the same load this run, and widening budgets without evidence would only mask real regressions; if another file flakes under load, it gets its own evidence-based bump.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}