[juansuaya17]

Claude custom agent definitions can override the parent or owner effort. Inheriting that value made a child thread display an effort it might never use. The adapter now records explicit launch effort and preserves it on subsequent updates; main's canonical model and effort validation remain intact.

The shared web/mobile native-subagent header also stops supplying catalog defaults or current values when the child has no selected or matching reported effort. Explicit and matching reported values, including the provider's Default sentinel, remain visible.

Rebuilt from @juansuaya17's #17042, preserving the contributor commits and merging main. The original model-copy helper was replaced with the current adapter's effort handling.

Verification

• Existing Claude adapter and shared header tests were updated rather than adding another test file. Baseline: 286 adapter tests and 44 shared tests; revised regressions exposed 2 adapter and 5 shared failures before implementation. After: 332 passed across both full suites (286 adapter, 46 shared).
• Scoped server and client-runtime typechecks passed. Changed-file lint and formatting passed, with one existing unused layer lint warning.
• One independent adversarial round: Claude Opus 5.5 and Codex gpt-6-sol found no code blockers across the complete diff against main.
• Live headless Linux web app, isolated SQLite fixtures: the same child with no effort displayed Medium using main's formatter and no effort using the head formatter; a separate child with explicit Low retained Low. These display fixtures do not launch a real Claude agent; SDK event tests cover parent/owner inheritance and later child updates.
• Chromium's system fonts measured zero width; the preview tab loaded a local DejaVu font for both captures. No product font changes.
• Not verified: a live Claude custom agent definition/account, macOS, Windows, iOS, Android, or Electron desktop. Mobile uses the shared formatter and was inspected, not launched.

Before — main formatter, unknown child effort:

After — head formatter, same child and viewport:

After — explicit effort remains visible:

Implementation: Codex gpt-6.1-sol; tests: Codex gpt-6-astra; reviews: Claude claude-opus-5-5 and Codex gpt-6-sol, through T3 Code delegated tasks. Maintainer integration: Codex gpt-6.1-sol in T3 Code.

[macroscopeapp]

Approvability

Verdict: Approved at cf65974

Macroscope's review found this PR approvable — Claude subagents will stop reporting their parent’s effort when their own effort is unknown. This is a narrow metadata correction that preserves explicit values without changing provider execution defaults.

Notes:

• No code objects were reviewed. Approvability was decided on eligibility alone.

^{You can add or adjust custom eligibility rules. Learn more.}

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

New subagents no longer inherit effort from their owner or the parent model selection. Existing subagents retain their recorded effort. Tests check explicit and unspecified effort values for nested subagents and their child threads.

Changes: Explicit model effort

Layer / File(s)	Summary
Record explicit subagent effort apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts, apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts, apps/server/src/orchestration-v2/testkit/OrchestratorReplayRecovery.integration.test.ts	New subagents record effort only when the launch input supplies it. Tests check explicit and unspecified effort values for nested subagents and their child threads. The recovery test no longer asserts that the child thread has effort: high.

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix

Possibly related issues

• pingdotgg/t3code#24143 — This issue describes parent effort appearing when a Claude subagent definition sets a different effort. This change removes the parent-effort fallback for new subagents.
• pingdotgg/t3code#29200 — This issue describes child threads displaying parent effort. This change leaves effort unset when a new subagent does not explicitly specify it.

Possibly related PRs

• pingdotgg/t3code#29237 — This PR changes Claude fork-skill subagents so they do not inherit parent effort.
• pingdotgg/t3code#29217 — This PR leaves fork effort unset instead of inheriting parent effort.
• pingdotgg/t3code#29248 — This PR suppresses inferred effort labels when a child model is unreported.

Suggested labels: bugfix

Suggested reviewers: rakshithbhat03

Merge Risk: Minimal · up to cf659

Claude subagents will no longer show the parent's effort when none was set explicitly, and explicitly set values are still shown. No concrete merge-blocking risk was identified in the supplied change.

Architecture risk: Low · up to dd9ea

The change affects 2 systems.

Changed systems: apps/server, packages/client-runtime

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — apps/server (service) was modified; 2 changed files map to changed impact.
• observed — packages/client-runtime (library) was modified; 2 changed files map to changed impact.

Before / after behavior

• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: The opus subagent case now expects no recorded effort when the Agent call omits it, rather than "max".
• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: The nested-subagent test is renamed to cover only explicit effort, and its unspecified nested subagent gets a distinct identifier instead of being described as inheriting effort.
• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: The test launches an unspecified nested subagent and an explicitly "max" subagent; it no longer frames the first call as an inheritance case.
• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: Expected effort for the unspecified nested subagent and its child thread changes from "low" to undefined; the outer "low" and explicit nested "max" expectations remain.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main change: Claude subagents no longer inherit or display the parent effort.
Description check		The description explains the problem, implementation, affected shared UI behavior, verification results, limitations, and review context. It does not use separate Problem, Change, or Scope and approva…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Description check

Explanation

The description explains the problem, implementation, affected shared UI behavior, verification results, limitations, and review context. It does not use separate Problem, Change, or Scope and approval headings, and it does not explicitly document maintainer approval, but the required information is mostly present.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: a03d9bc
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: bb0ce7d7-f2ac-4d87-b567-04fa67b0d2ec

Commits

Reviewing files that changed from the base of the PR and between b66dfea and cf65974.

Files selected for processing: 1

• apps/server/src/orchestration-v2/testkit/OrchestratorReplayRecovery.integration.test.ts

Files with no reviewable changes: 1

• apps/server/src/orchestration-v2/testkit/OrchestratorReplayRecovery.integration.test.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

Dismissing prior approval to re-evaluate 9a12353

[juliusmarminge]

Taking over this one for batch claude-01. I’ll update the fix against current main while preserving explicit subagent effort handling, keep focused regression coverage, and verify it before publishing.

Dismissing prior approval to re-evaluate dd9eaa6

[saphid]

Source review of dd9eaa6: an unsupported stored effort can still turn into a catalog-default label. In formatModelSelectionEffort, a saved { id: "effort", value: "old-level" } passes the new string-presence guard. However, getProviderOptionDescriptors has already normalized that obsolete choice through resolveDescriptorChoiceValue, which falls back to descriptor currentValue or the catalog default. getProviderOptionCurrentLabel then displays that fallback, e.g. Medium, although neither the child nor provider reported Medium.

Please add an obsolete stored-choice regression with a descriptor default/current value, and validate the raw stated choice against the offered options before displaying its label. Preserve the supported reported Default sentinel. A matching reported obsolete value follows a different raw-value path, so I am not asserting the same fallback for that case.

Exact-head source: header guard, descriptor normalization. This is independently checked source reasoning, not an executed test or real Claude reproduction. Exact-base/head custom-agent and client verification remain pending; the supplied display fixtures do not establish the provider's actual effort.

[smolpaw]

#24143 has a live repro of this on 0.1.2 with a real Claude custom agent definition (effort: low, thread at xhigh). T3 stores xhigh, and Claude Code runs the subagent at low. The steps are runnable by an agent and check both sides: the T3 state database and the subagent's Claude Code transcript.

It also covers a second path: a PreToolUse hook that sets effort through updatedInput gets the same wrong label.

Dismissing prior approval to re-evaluate cf65974

[juliusmarminge]

Evidence

On both commits, configure the custom effort-probe agent with effort: low, start a fresh Claude parent at Extra High, and request a native Agent call with subagent_type: effort-probe, prompt Reply OK, and no effort parameter. Open the completed child. Both native child transcripts report low; base saves and displays Extra High, while head leaves effort absent.

Environment for both sides: real web client and server, isolated fresh databases, identical synthetic project, Chromium at 1280×800, light theme. Images crop the sidebar; black masks redact private model identifiers. Recordings show opening the completed child from its parent, with two seconds of final-frame hold added.

Before	After
a03d9bce02	cf6597442f
Before recording	After recording
Before saved selection and native transcript excerpt	After saved selection and native transcript excerpt

Executed synthetic seed script · Read-only evidence extraction script

AI-assisted; verified with the tests and tools listed above. No tests were used as proof.

Not shown:

• Explicitly supplied Agent effort, nested subagents, and resumed subagents were not exercised; this proof covers the reported custom-agent call with omitted effort.