feat(sdd-execute): raise per-run wall-clock budget to 60m - #306
Conversation
All three execute tiers (haiku/sonnet/opus) used gh-aw's default 20-minute agent-job timeout. That budget is consumed by file exploration plus the in-sandbox `cargo build/test` pre-PR gate (step 6); a broad multi-file change (e.g. a workspace-wide caller migration) cannot finish a cold compile + tests within it, so the agent job is killed at 20m with no PR — a clean wall-clock timeout, not a turn or capability limit. Opus is slower wall-clock than sonnet and hit this hardest. Set `timeout-minutes: 60` in each execute .md frontmatter and recompile the locks (patched gh-aw v0.77.5, byte-identical to CI). Only the agent job's `timeout-minutes` and `GH_AW_TIMEOUT_MINUTES` change (20 → 60); the activation and conclusion jobs keep their own short timeouts. Consumers calling these reusable locks @main pick up the new budget on the next run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ss8WXtUUr9PBHuJcJMseBj
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (6)
📝 WalkthroughSummary by CodeRabbit
WalkthroughThe ChangesSDD Execute Workflow Timeout Increase (all tiers)
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Raise the agent-job wall-clock budget on all three execute tiers (haiku/sonnet/opus) from gh-aw's default 20m → 60m.
Why
The 20m budget is spent on file exploration + the in-sandbox
cargo build/testpre-PR gate (step 6). A broad multi-file change (e.g. a workspace-wide caller migration) cannot finish a cold compile + tests within it, so the agent job is killed at 20m with no PR — a clean wall-clock timeout, not a turn/capability limit.Observed on a consumer pilot run: a
NetworkModeenum + 7-file caller migration timed out at ~20m across three attempts (sonnet ×3). Bumping the model does not help a time limit (opus is slower wall-clock); extending the budget does.Change
timeout-minutes: 60added to eachsdd-execute-{haiku,sonnet,opus}.mdfrontmattertimeout-minutesandGH_AW_TIMEOUT_MINUTESchange (20 → 60). Activation/conclusion jobs keep their own short timeouts. Digest snapshot unchanged.Propagation
Consumers call these reusable locks
@main, so the new budget takes effect on the next execute run after merge. GitHub's per-job ceiling is 360m; 60m stays well within it.🤖 Generated with Claude Code