Repository navigation
fix(deps): align agent-runtime and agent-interface with agent-eval 0.221.3 - #229
Conversation
…ess-sessions 0.2.0 with agent-eval 0.221.3
agent-eval 0.221.3 beside agent-interface 5.2.0 and agent-runtime 0.318.0 loaded a second agent-interface and agent-core graph: a tiny facts call took 25% more CPU (1.55 to 1.95 s on beelink1), enough to push CLI-spawning tests past 30 s on a loaded runner. Aligned, it is back to 1.7 s.
tangletools
left a comment
There was a problem hiding this comment.
✅ Auto-approved PR — 9d45f371
Blanket team auto-approval is intentional. This is not a code review.
No automated review runs on this PR. This approval rests on the rule above alone.
tangletools · auto-approval · reason: blanket_auto_approve · 2026-10-25T00:41:00Z
tangletools
left a comment
There was a problem hiding this comment.
✅ Auto-approved PR — ac53b7ae
Blanket team auto-approval is intentional. This is not a code review.
No automated review runs on this PR. This approval rests on the rule above alone.
tangletools · auto-approval · reason: blanket_auto_approve · 2026-10-25T00:47:04Z
tangletools
left a comment
There was a problem hiding this comment.
✅ Auto-approved PR — 8df9d440
Blanket team auto-approval is intentional. This is not a code review.
No automated review runs on this PR. This approval rests on the rule above alone.
tangletools · auto-approval · reason: blanket_auto_approve · 2026-10-25T00:48:59Z
|
Gate at 8df9d44 on beelink1-wsl, run directly in a clean clone at this SHA. beelink-gate held on CPU pressure: load average 131–142, PSI ~80.
|
Main has been red since #224/#225: CLI-spawning tests time out at 30 s.
Cause. #224 moved agent-eval to 0.221.3, whose agent-core 0.13.2 and agent-interface 5.7.1 (zod 4.6.6) loaded beside traces' own agent-interface 5.2.0 and agent-runtime 0.318.0 (agent-core 0.11, zod 4.6.5). Every CLI start compiled both graphs. On beelink1, a tiny
factscall went from 1.55 to 1.95 CPU-seconds (+25%), withcompileSourceTextModule495 → 755 ms and zod 300 → 683 ms in a CPU profile. At runner load that pushed the spawning tests past 30 s. Same six files at the same load (PSI ~83): b3d8302 105 s, 81/81 passed; main 153 s, 5 failed. No hang: each failing test passes alone.Fix. Align the direct deps on agent-eval 0.221.3's stack: agent-runtime 0.348.1 and agent-interface 5.7.1. Typecheck is clean with no code change.
factsCPU is back to parity with b3d8302 (≈1.75 s both, four runs each); the six files run 81/81 in 71 s.harness-sessions stays at 0.1.7: 0.2.0's import holds 17.5 MB of heap vs 5.5 MB, which ran the 64 MB bounded-heap Codex rollout test out of memory. With 0.1.7 the Codex adapter's import heap is 12.2 MB (b3d8302: 10.9 MB).
Gate (beelink1, run directly at this SHA because beelink-gate is holding on CPU pressure) follows in a comment.
Run ID: run_66835efe8253ce375017d831711d5b00