feat(goose): save a Thinking level and match the Claude/Codex profile by salman1993 · Pull Request #1250 · block/buzz-app

salman1993 · 2026-10-16T15:09:06Z

Summary

goose agents can now save a Thinking level, as Claude Code can since #1089. goose itself supplies the levels: each one comes from the thought_level config option of a goose session opened on the agent's provider and model. Nothing is hardcoded in Buzz.

The goose agent profile also now matches Claude Code and Codex.

Thinking level

GooseSetup.efforts(choice) gets the levels from a short-lived session with no tools. It sends session/new, then set_config_option model when a model is set, then deletes the session. The result is cached per provider and model, and the cache is cleared when a provider is saved.
GooseSession.select() sets thinking_effort after the provider and model. goose keeps the level across model and provider changes.
goose has no value that means "use goose's default". So when an agent goes back to the default and its live session had a level set, the next turn reopens the session (stale() in sessions.ts).
The picker labels the default Codex-style. In the list it reads "Medium · Default"; when closed it reads "Medium". The default comes from goose's GOOSE_THINKING_EFFORT.
If the chosen model doesn't offer the saved level, the picker shows it as "(unavailable)" instead of silently dropping it.

Profile layout and labels

The Settings tab uses splitSettings, the same as Claude and Codex.
- Identity is a grouped card with Harness, Provider, Model and Thinking, followed by instructions.
- Runtime holds the provider keys with Test connection, the workspace, and sessions.
The agent-type row "Provider" is renamed to Harness (Claude, Codex, Goose). goose's "Model provider" is now Provider.
The New agent dialog keeps keys and Test connection next to the provider.

Testing

All goose tests plus the agent-related bundles pass (1,040 tests). tsc, biome lint and design:check are clean.
New coverage:
- setup.test.ts: the efforts() read, cache, and cleanup.
- sessions.test.ts: the level is applied, and the session reopens on the default.
- runtime.test.ts: the level is part of the agent's choice.
- tabs.test.tsx: saving a level, an unavailable level, and the Identity/Runtime split.
goose.live.test.ts: a live thinking-level test passed against goose-acp 1.50.

buzz-review-completed

🤖 Generated with Claude Code

goose agents can now keep a Thinking level. The levels come from the config options of a goose session on the agent's provider and model. The level is applied with session/set_config_option thinking_effort, and a session that had a level set is reopened when the agent goes back to goose's own default. The goose Settings tab now uses the split Identity/Runtime layout that Claude and Codex use. Identity is a grouped card with Harness, Provider, Model and Thinking. Provider keys, the workspace and sessions are under Runtime. The agent type row is renamed to "Harness", and goose's "Model provider" is now "Provider". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Salman Mohammed <smohammed@squareup.com>

salman1993 · 2026-10-16T15:10:17Z

chatgpt-codex-connector · 2026-10-16T15:10:57Z

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-16T15:16:22.970836Z	`beebcaa`	Draft marked ready

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.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: beebcaa806

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".

chatgpt-codex-connector · 2026-10-16T15:16:19Z

+      live.fingerprint !== this.options.fingerprint() ||
+      Boolean(live.session.effort && !this.options.choice().effort)


Reopen before running a queued default-effort turn

When the owner clears Thinking while a session is busy, reconfigure() leaves the live session open; if another message specifying a different per-message model arrives before the first turn finishes, send() queues it instead of steering it. select() cannot unset the session's previous chosen.effort, so that queued next turn runs at the old explicit level even though the saved setting now says to use goose's default. Ensure a pending default reset makes queued turns wait for a reopened session.

Useful? React with 👍 / 👎.

🤖 Leaving this as is, deliberately. The race is real but narrow: Thinking has to be cleared mid-turn, and a message has to be queued rather than steered, which only happens when it asks for a different model or arrives as the turn ends. At worst one turn runs at the previous level, and the session closes once it falls idle.

The settings that need a new session (workspace, instructions, scope, team, name in fingerprint()) already work this way: a busy session keeps running its queued turns and closes when it next falls idle, as reconfigure() documents. Making queued turns wait for a reopened session would add machinery, and if we want it, it should cover those settings too, in a separate change.

chatgpt-codex-connector · 2026-10-16T15:16:19Z

+    const [listing, reload] = useRead(() => setup.providers(), "providers");
+    const id = choice.provider || listing.value?.defaults.provider;
+    const provider = listing.value?.providers.find((item) => item.id === id);


Restore provider discovery error recovery in Runtime

In the new Runtime section, when setup.providers() rejects—for example because goose transiently fails to start—listing.status is ignored and provider remains undefined, so all provider key/sign-in UI silently disappears while only Test connection remains, with no error or retry action. The previous ProviderFields path surfaced the failure and a Try again button; restore equivalent scoped recovery here as required by the states guidance in src/shared/design-system/DESIGN.md:220-228.

AGENTS.md reference: AGENTS.md:L46-L51

Useful? React with 👍 / 👎.

🤖 Fixed in 300748a. When goose cannot list providers, ProviderSetup now shows the error next to a Try again button that reads the list again, the same pattern ProviderKeys uses. The new test "says in Settings when goose cannot list providers, and tries again" covers it. It fails without the fix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Salman Mohammed <smohammed@squareup.com>

The Runtime section dropped the provider's keys without a word when goose failed to list providers. It now shows the error and Try again, as the combined view did before the split. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Salman Mohammed <smohammed@squareup.com>

kalvinnchau

🤖 Approval submitted at the user's direction for commit 300748a.

Review: the documentation-only delta at 08ea6b4 introduced no new actionable findings. The latest delta adds the missing Runtime provider-discovery error and retry UI, with a regression test; source inspection confirms the error state exposes the existing reload action.

Remaining findings from the earlier review: queued Default reset can miss the next turn, and effort-discovery RPC/process lifetime remains unbounded. This approval does not assert those issues are resolved.

Validation: prior independent static, security and integration source checks; latest delta inspected, new test not run. Browser/native workflows remain unverified because permitted web startup requires Keychain access and Goose registration requires Tauri.

Main fails typecheck since #1250 made effort required and #1185 added this fixture without it. Same one-line fix as in #1257 and #1260. Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz> Signed-off-by: Logan Johnson <loganj@squareup.com>

…ui-polish * buzz-app/main: test(agents): fix main typecheck — add effort to Goose fixture (#1262) fix(messages): reject unclaimed file and link drops app-wide (#1075) feat(channels): create and read system channels (#965) Refine message layout and bubble reading widths (#1218) feat(goose): save a Thinking level and match the Claude/Codex profile (#1250) Signed-off-by: Taylor Ho <taylorkmho@gmail.com>

salman1993 marked this pull request as ready for review October 16, 2026 15:10

salman1993 requested a review from a team as a code owner October 16, 2026 15:10

chatgpt-codex-connector Bot reviewed Oct 16, 2026

View reviewed changes

salman1993 and others added 2 commits October 16, 2026 11:39

docs(goose): explain why thinking levels need a throwaway session

08ea6b4

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Salman Mohammed <smohammed@squareup.com>

kalvinnchau approved these changes Oct 16, 2026

View reviewed changes

salman1993 mentioned this pull request Oct 16, 2026

feat(pi): add the Pi agent type on Pi durable #1103

Closed

salman1993 merged commit 1228d79 into main Oct 16, 2026
32 checks passed

salman1993 deleted the claude/goose-thinking-level-parity-7a1dbc branch October 16, 2026 16:07

salman1993 mentioned this pull request Oct 16, 2026

feat(goose): show goose and provider marks on agent cards #1260

Merged

loganj mentioned this pull request Oct 16, 2026

test(agents): fix main typecheck in Goose launch tests #1261

Closed

kalvinnchau mentioned this pull request Oct 16, 2026

test(agents): fix main typecheck — add effort to Goose fixture #1262

Merged

salman1993 mentioned this pull request Oct 16, 2026

fix(agents): add effort to launch-boundary Goose test fixture #1263

Closed

klopez4212 mentioned this pull request Oct 16, 2026

Improve AI defaults and add Goose configuration #1266

Merged

loganj mentioned this pull request Oct 16, 2026

fix(messages): stop the channel timeline jumping back when agents or search read old messages #1256

Merged

Navigation Menu

Repository navigation

feat(goose): save a Thinking level and match the Claude/Codex profile - #1250

feat(goose): save a Thinking level and match the Claude/Codex profile#1250
salman1993 merged 3 commits into
mainfrom
claude/goose-thinking-level-parity-7a1dbc

salman1993 commented Oct 16, 2026 •

edited

Loading

Uh oh!

salman1993 commented Oct 16, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 16, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Oct 16, 2026

Uh oh!

salman1993 Oct 16, 2026

Uh oh!

chatgpt-codex-connector Bot Oct 16, 2026

Uh oh!

salman1993 Oct 16, 2026

Uh oh!

kalvinnchau left a comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		live.fingerprint !== this.options.fingerprint() \|\|
		Boolean(live.session.effort && !this.options.choice().effort)

Conversation

salman1993 commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Thinking level

Profile layout and labels

Testing

Uh oh!

salman1993 commented Oct 16, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

salman1993 Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

salman1993 Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

kalvinnchau left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

salman1993 commented Oct 16, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Oct 16, 2026 •

edited

Loading