PR #70793: fix: wizard no clobber model.primary on re-run

• Repository: openclaw/openclaw
• Author: Sathvik-1007
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/70793

Description (problem / solution / changelog)

Summary

• Problem: Configure wizard silently overwrites agents.defaults.model.primary when user re-runs model section. Two separate code paths clobber the user's chosen primary model.
• Why it matters: User explicitly picks a primary model via CLI, re-runs wizard for provider auth, and their choice is silently replaced — no warning, no indication.
• What changed: Two one-line fixes: (1) applyDefaultModel() now preserves existing primary instead of unconditionally overwriting; (2) applyModelFallbacksFromSelection() no longer injects a phantom primary when none was configured.
• What did NOT change (scope boundary): No changes to wizard flow, provider auth logic, model resolution, or config schema. Only the two assignment expressions that clobber model.primary.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #70696
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause (Bug 1): applyDefaultModel() in src/plugins/provider-auth-choice-helpers.ts:138 sets primary: model unconditionally. When wizard re-runs with setDefaultModel: true, the provider's default model (e.g. openrouter/auto) overwrites whatever the user had previously configured.
• Root cause (Bug 2): applyModelFallbacksFromSelection() in src/flows/model-picker.ts:833 uses primary: existingPrimary ?? resolvedKey. When no primary was configured, existingPrimary is undefined and resolvedKey (the hardcoded default) gets injected as a phantom primary the user never chose.
• Missing detection / guardrail: No test asserted that existing primary survives re-application of default model or that absent primary stays absent after fallback selection.
• Contributing context: The restoreConfiguredPrimaryModel helper at line 299 of provider-auth-choice.ts already handles the setDefaultModel: false path correctly — the setDefaultModel: true path was simply missed.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file:
  • src/plugins/provider-auth-choice-helpers.test.ts — 4 new tests for applyDefaultModel
  • src/commands/model-picker.test.ts — 1 new test for phantom primary injection
• Scenario the test should lock in:
  • applyDefaultModel preserves an existing primary instead of overwriting
  • applyDefaultModel preserves existing fallbacks alongside existing primary
  • applyModelFallbacksFromSelection does not inject phantom primary when none configured
• Why this is the smallest reliable guardrail: Both bugs are pure data transformation (config in → config out) with no side effects, so unit tests on the helper functions are sufficient.
• Existing test that already covers this (if any): None — these helpers had no applyDefaultModel tests and no phantom-primary assertion.

User-visible / Behavior Changes

• User's configured agents.defaults.model.primary is no longer silently overwritten when re-running the configure wizard's provider auth section.
• When no primary model is configured, the fallback selection flow no longer injects a default model as primary.

Diagram (if applicable)

Before (Bug 1):
[wizard re-run] -> applyDefaultModel(config, "openrouter/auto")
  -> primary: "openrouter/auto"  (CLOBBERS user's "anthropic/claude-opus-4-6")

After (Bug 1):
[wizard re-run] -> applyDefaultModel(config, "openrouter/auto")
  -> primary: "anthropic/claude-opus-4-6"  (PRESERVED, provider default ignored)

Before (Bug 2):
[no primary configured] -> applyModelFallbacksFromSelection(config, selections)
  -> { primary: "openai/gpt-5.5", fallbacks: [...] }  (PHANTOM primary injected)

After (Bug 2):
[no primary configured] -> applyModelFallbacksFromSelection(config, selections)
  -> { fallbacks: [...] }  (no primary key — correct)

Security Impact (required)

• New permissions/capabilities? No
• Secrets/tokens handling changed? No
• New/changed network calls? No
• Command/tool execution surface changed? No
• Data access scope changed? No

Repro + Verification

Environment

• OS: Linux (CachyOS/Arch)
• Runtime/container: Node 22+, pnpm
• Model/provider: Any (reproduces with OpenRouter, OpenAI)
• Relevant config: agents.defaults.model.primary set to any non-default value

Steps

1. Run openclaw configure wizard, set a primary model (e.g. anthropic/claude-opus-4-6)
2. Re-run wizard, go through provider auth section (triggers applyDefaultModel with provider's default)
3. Check ~/.openclaw/default.yaml — agents.defaults.model.primary is now the provider's default, not user's choice

Expected

• agents.defaults.model.primary remains anthropic/claude-opus-4-6

Actual

• agents.defaults.model.primary silently changed to provider default (e.g. openrouter/auto)

Evidence

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

5 new tests added covering both bugs. All 22 tests pass in affected test files. 0 lint errors on all 4 changed files.

Human Verification (required)

• Verified scenarios: Both fixes verified via unit tests — applyDefaultModel preserves existing primary, applyModelFallbacksFromSelection does not inject phantom primary. Full test suite (22/22) passes. Typecheck via pnpm check:changed passes. Lint (oxlint) clean on all 4 files.
• Edge cases checked: No existing primary (should set it), existing primary with fallbacks (should preserve both), string-form existing model config (should extract primary correctly).
• What you did not verify: Full e2e wizard flow on a live instance (no connected provider accounts in dev env).

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

• Backward compatible? Yes
• Config/env changes? No
• Migration needed? No

Risks and Mitigations

• Risk: Edge case where existingModel is a plain string (legacy config format) and extraction of existingPrimary fails silently.
  • Mitigation: applyDefaultModel now handles both string and object model formats via explicit type checks (lines 126–132). Covered by test.

AI-assisted PR (OpenCode + Claude). Fully tested. I understand what the code does — traced both bugs through the full wizard flow before writing fixes.

Changed files

• src/commands/model-picker.test.ts (modified, +17/-0)
• src/flows/model-picker.ts (modified, +1/-1)
• src/plugins/provider-auth-choice-helpers.test.ts (modified, +54/-1)
• src/plugins/provider-auth-choice-helpers.ts (modified, +7/-1)

PR #3: fix(configure): preserve custom primary model when reconfiguring auth

• Repository: suboss87/openclaw
• Author: suboss87
• State: open | merged: False
• Link: https://github.com/suboss87/openclaw/pull/3

Description (problem / solution / changelog)

Summary

• openclaw configure --section model (or openclaw configure > Model section) silently resets agents.defaults.model.primary to the provider's default (e.g. openrouter/auto) even when the user has previously set a custom primary model via openclaw config set.
• Root cause: applyDefaultModelChoice with setDefaultModel: true unconditionally called applyDefaultConfig, which calls applyAgentDefaultModelPrimary on every provider handler, overwriting whatever primary was already in config.
• Fix: check for an existing primary before branching. If one is set, call applyProviderConfig instead (sets up auth/allowlist without touching primary). Only call applyDefaultConfig on a fresh setup where no primary has been configured yet.

Root Cause

src/commands/auth-choice.default-model.ts:applyDefaultModelChoice - the setDefaultModel: true branch:

// before
if (params.setDefaultModel) {
  const next = params.applyDefaultConfig(params.config);  // always clobbers primary
  ...
}

Every provider's applyDefaultConfig (e.g. applyOpenrouterConfig) ends with applyAgentDefaultModelPrimary(cfg, PROVIDER_DEFAULT_MODEL), which unconditionally writes the provider default into agents.defaults.model.primary. On a first-time run this is correct. On a reconfigure run it clobbers any custom value the user set via CLI.

Fix

src/commands/auth-choice.default-model.ts - 12 lines added:

const existingPrimary = resolveAgentModelPrimaryValue(params.config.agents?.defaults?.model);
if (existingPrimary) {
  const next = params.applyProviderConfig(params.config);
  return { config: next };
}

applyProviderConfig (e.g. applyOpenrouterProviderConfig) adds the provider model to the allowlist without touching primary.

Test Plan

• 4 new unit tests in src/commands/auth-choice.default-model.test.ts:
  • fresh install (no existing primary) - sets provider default, shows note
  • existing object-form primary - preserved, applyDefaultConfig NOT called (regression for #70696)
  • existing string-form primary - preserved
  • setDefaultModel: false path - unchanged behavior, agentModelOverride returned
• All 4 tests pass
• Existing model-picker.test.ts (7 tests) still pass
• Lint + format clean

Closes #70696

Generated by Claude Code

Changed files

• src/commands/auth-choice.default-model.test.ts (added, +141/-0)
• src/commands/auth-choice.default-model.ts (modified, +12/-0)