fix: [ENG-2469] update Google default model to gemini-3-flash-preview

[cuongdo-byterover]

The previous default gemini-2.5-flash returns thinking-only responses when tools are present, which the agent loop cannot consume — every curate fails validation immediately.

Switch the default surfaced by both the provider registry and the Google provider config to gemini-3-flash-preview, matching the docs recommendation list.

Summary

• Problem:
• Why it matters:
• What changed:
• What did NOT change (scope boundary):

Type of change

• Bug fix
• New feature
• Refactor (no behavior change)
• Documentation
• Test
• Chore (build, dependencies, CI)

Scope (select all touched areas)

• TUI / REPL
• Agent / Tools
• LLM Providers
• Server / Daemon
• Shared (constants, types, transport events)
• CLI Commands (oclif)
• Hub / Connectors
• Cloud Sync
• CI/CD / Infra

Linked issues

• Closes #
• Related #

Root cause (bug fixes only, otherwise write N/A)

• Root cause:
• Why this was not caught earlier:

Test plan

• Coverage added:
  • Unit test
  • Integration test
  • Manual verification only
• Test file(s):
• Key scenario(s) covered:

User-visible changes

List user-visible changes (including defaults, config, or CLI output).
If none, write None.

Evidence

Attach at least one:

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording

Checklist

• Tests added or updated and passing (npm test)
• Lint passes (npm run lint)
• Type check passes (npm run typecheck)
• Build succeeds (npm run build)
• Commits follow Conventional Commits format
• Documentation updated (if applicable)
• No breaking changes (or clearly documented above)
• Branch is up to date with main

Risks and mitigations

List real risks for this PR. If none, write None.

• Risk:
  • Mitigation:

[github-actions]

Claude finished @cuongdo-byterover's task in 1m 19s —— View job

---

Code Review

• Gather context and read changed files
• Analyze changes for correctness, security, and code quality
• Post review feedback

---

Summary

Overall score: 4/5

The change is minimal, targeted, and correct. gemini-3-flash-preview is already used as the default in several other parts of the codebase (src/server/constants.ts, src/agent/infra/llm/providers/byterover.ts, src/agent/infra/agent/interactive-loop.ts, etc.), making this update consistent. Two locations are correctly brought into alignment.

---

Observations

praise(correctness): Both default model fields updated consistently

src/agent/infra/llm/providers/google.ts:24 and src/server/core/domain/entities/provider-registry.ts:164 both changed in lockstep. No stale reference was left behind in these two files, and the new model is already validated in the model registry at src/agent/core/domain/llm/registry.ts:160.

---

issue(testing): No tests added — violates mandatory TDD policy

CLAUDE.md requires TDD: write failing tests first, confirm they fail, then implement. This PR skips that step entirely. There is no test asserting that googleProvider.defaultModel === 'gemini-3-flash-preview' or that PROVIDER_REGISTRY.google.defaultModel === 'gemini-3-flash-preview'. If someone reverts this value in a future refactor, nothing will catch it.

Suggested additions:

• A unit test in test/unit/agent/llm/providers/ asserting googleProvider.defaultModel
• A unit test in test/unit/ asserting PROVIDER_REGISTRY.google.defaultModel

Fix this →

---

nitpick(docs): PR description template is largely unfilled

The Summary bullet points, Root cause fields, Test plan, Evidence, and Checklist are all left as template placeholders. For a fix that has a concrete root cause and a clear user-visible change (Google provider now defaults to a working model), these sections are easy to fill in and help reviewers and future bisect readers understand the change.

---

note(scope): gemini-2.5-flash references in tests are fine

Tests under test/unit/agent/llm/tokenizers/, test/unit/agent/llm/retry/, etc. still use 'gemini-2.5-flash' as a concrete model name to test other logic (tokenizer behavior, retry policy). These do not need to change — gemini-2.5-flash remains a valid registered model; it just is no longer the default.

---

note(no security concerns)

No security implications. This is a pure string constant change affecting model selection defaults.

[bao-byterover]

LGTM