feat(fork): promote v0.0.36 Dulli rebuild

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

What Changed

This promotes the reviewed T3 Dulli rebuild based on upstream stable v0.0.36.

• Preserves the Dulli desktop/mobile identity, installation paths, update repository, signing continuity, Android package, and @yjosh/t3 CLI.
• Integrates Pi 0.84.4 as a first-class provider with profiles, usage, task controls, reasoning, tools, background terminals, durable child transcripts, subagents, and workflows.
• Restores Workler-backed workspace creation while retaining upstream Start from origin behavior.
• Restores the dedicated Fork features settings section and Dulli workspace/branch defaults.
• Keeps fork release automation isolated in .github/workflows/fork-desktop-release.yml; upstream publishers, relay deployment, EAS, and store workflows remain disabled.
• Rebuilds the web/desktop Agents inspector around one workflow roster item, a phase tree, per-agent model labels, and live chat-style agent transcripts.

Promotion note: this head records the previous main as a merge parent while preserving the rebuilt tree exactly. Merge with a merge commit, not squash or rebase. The previous fork main is also preserved at archive/pre-v0.0.36-dulli-rebuild.

Why

The previous fork main predated upstream stable v0.0.36 and carried provider, workspace, and release implementations that no longer matched the canonical runtime. A normal content merge would reintroduce obsolete architecture. This promotion keeps the reviewed rebuild as the exact resulting tree while preserving both histories.

UI Changes

Fork settings

Before: fork controls mixed into General	After: dedicated fork section and defaults

Workflow details

Before: one aggregate model label	After: model shown on each workflow agent

Verification

• Focused server tests passed for Pi projection/replay, runtime ingestion, task controls, settings persistence, Workler behavior, release helpers, and publisher isolation.
• Focused web/client/contracts tests passed for agent transcripts, workflows, settings, trait scoping, composer behavior, and runtime folding.
• Web, server, client-runtime, and contracts typechecks passed.
• Targeted lint, formatting, and diff checks passed.
• Playwright verified workflow navigation, per-agent models, persisted child conversations, prompt deduplication, responsive layouts, and fork settings with no new console errors.
• The release path was exercised by v0.0.36-pi.2, including desktop matrices, updater manifests, signatures, Android identity/signing, GitHub assets, and npm publication. Follow-up UI commits remain unreleased.
• Live mobile verification remains pending because this host has no available device or emulator; the current inspector redesign does not change mobile behavior.

Checklist

• This PR is focused on one concern: promoting the clean Dulli rebuild.
• I explained what changed and why.
• I included before/after screenshots for UI changes.
• No new motion or animation requires a video.

Implemented with GPT-5.6 Sol using the pi coding-agent harness.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed2026-09-02T19:35:00.711115Z	f3cf060	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.3%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+459 B (+7.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.4 KiB	55.5 KiB	+4.1 KiB (+8.0%)	66.4 KiB	✅
Codex	Live turn messages	16	9	−7 (−43.8%)	21	✅
Claude	Total thread wire	11.5 KiB	13.3 KiB	+1.8 KiB (+15.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.6 KiB	6.9 KiB	+1.3 KiB (+23.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	6.4 KiB	+491 B (+8.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.2 KiB	56.4 KiB	+4.1 KiB (+7.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 33b5e46 · PR result: f3cf060 · Source CI: success

Scenario and decoded snapshot size

10 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.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[chatgpt-codex-connector]

💡 Codex Review

t3code/apps/server/src/provider/Layers/PiProvider.ts

Lines 245 to 248 in 271f05f

	constdiscovery=yield*discoverPiModels({
	agentDir: settings.agentDir||undefined,
	profile: settings.profile||undefined,
	}).pipe(

Apply the instance environment during Pi model discovery

When a Pi provider instance supplies credentials or custom-provider configuration through its Environment settings, PiDriver passes that merged environment to the version probe and spawned Pi process, but this SDK discovery runs in the server process without applying it. Consequently getAvailable() can report no models and an unauthenticated provider even though the actual Pi subprocess has valid credentials, leaving the model picker unusable for that supported configuration. Run discovery with the same per-instance environment.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the Pi model-discovery finding in 9f1025404: provider status now passes the merged instance environment into SDK discovery, which runs in a bounded worker so extension/provider credentials never mutate the server process environment. Added a real extension-backed isolation test.

The same commit addresses the multi-instance Usage finding and the failing compatibility-boundary test. Focused verification: 52 tests passed, affected contracts/shared/server typechecks passed, targeted format/lint passed, and the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9f1025404e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed both follow-up findings in 68b56ea32. Usage now resolves roots from the hydrated Claude, Codex, Grok, and Pi instance set with per-instance environments, while incomplete directory/file reads surface partial coverage and cannot prune cache entries from incomplete walks.

Focused verification: 60 tests passed across provider discovery, Usage resolution/reading/aggregation, and shared merging; affected typechecks and targeted format/lint passed; the server bundle build passed.

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:68b56ea322

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[YJJosh]

Addressed the complete-vs-partial source ordering finding in f3cf060a1, with focused regression coverage. The previous 68b56ea32 CI run was fully green; the latest commit is a two-file shared merge/test change and its focused tests, typecheck, format, and lint pass.

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

Reviewed commit:f3cf060a1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".