Repository navigation
fix(server): remove one-child-per-model instruction - #25359
Conversation
commented
Oct 26, 2026
ApprovabilityVerdict: Needs Human Approval Macroscope's review found this PR requires human approval — This removes a constraint from the server's default orchestration prompt, allowing agents to delegate multiple children to the same model on an existing production path. The change is small but alters default product behavior and warrants human review. You can add or adjust custom eligibility rules. Learn more. |
commented
Oct 26, 2026
|
No actionable comments were generated in the recent review.
|
| Layer / File(s) | Summary |
|---|---|
Delegation instruction update packages/provider-core/src/server/orchestrationInstructions.ts |
The instructions remove the requirement to use one child per named model. Other delegation instructions remain. |
Priority:
Estimated code review effort: 1 (Trivial) | ~3 minutes
Change: Bug fix · Severity of issue fixed: Low
Possibly related issues
pingdotgg/t3code#24967reports that the removed restriction blocked requests for two subagents on the same model. This change removes that restriction; it does not add the proposed orchestration eval.
Possibly related PRs
- pingdotgg/t3code#25154: This PR also changes the shared instruction that limited delegation to one child per model.
Suggested labels: bugfix
Suggested reviewers: t3dotgg
Merge Risk: 
Minimal · up to 4f002
-
No actionable merge-blocking risk was identified.
Pre-merge checks: 
4
Finishing Touches
-
📝 Generate docstrings
-
- Commit to this branch
- Create a new PR
🧪 Generate unit tests (beta)
-
- Commit to this branch
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Comment @coderabbitai help to get the list of available commands.
commented
Oct 26, 2026
Thread transfer impact✅ Thread transfer remains within every enforced ceiling.
Baseline: Scenario and decoded snapshot size10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed. |
Problem
The injected "One child per named model" instruction makes agents refuse requests for multiple subagents using the same model. For example, a request for 20 Sol workers and 20 Luna workers was reduced to one worker per model.
Change
Remove that sentence from the shared orchestration prompt so it no longer imposes a per-model child limit across providers and clients.
Scope and approval
This is a focused fix for an obvious prompt bug in the existing delegation capability. Related to #24967 and the existing fix proposed in #25154.
Verification
vp test run apps/server/src/provider/CodexDeveloperInstructions.test.ts: all five tests passed, including instruction routing and the per-entry byte limit.vp run check:changed: formatting, lint, both knip passes, and typechecks for 21 affected packages passed. Two existing looping.envsymlinks were temporarily moved for this check and restored afterward.Model: GPT-6.1 Sol. Harness: Codex.