fix(hooks): use configured primary model for slug generation (#14272)

[lailoo]

Summary

Fixes #14272

Problem

generateSlugViaLLM in src/hooks/llm-slug-generator.ts hardcodes anthropic as the provider and claude-haiku as the model for slug generation. Users who configure a different primary model (e.g. openai/gpt-4.1-mini) still get Anthropic API calls, which fail if they don't have an Anthropic API key.

Fix

Use resolveConfiguredModelRef() to read the user's configured primary model and pass its provider/model to runEmbeddedPiAgent. Falls back to the original Anthropic defaults when no model is configured.

Reproduction & Verification

Before fix (main branch) — Bug confirmed:

• src/hooks/llm-slug-generator.ts hardcodes provider: "anthropic" and modelId: "claude-haiku" — no test file exists on main.
• Users with non-Anthropic configs get FailoverError when session-memory hook tries to generate slugs.

After fix — All verified:

✓ passes configured primary model to runEmbeddedPiAgent (#14272)
✓ falls back to default Anthropic model when no primary is configured
✓ returns cleaned slug from LLM response
✓ returns null when LLM returns no payloads
✓ returns null when LLM throws
5 tests pass (pnpm vitest run src/hooks/llm-slug-generator.test.ts)

Testing

• ✅ 5 tests pass (pnpm vitest run src/hooks/llm-slug-generator.test.ts)
• ✅ Lint passes

[greptile-apps]

_{3 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Mocks not reset

mockRunEmbeddedPiAgent is shared across all tests and its call history is never cleared, so assertions like toHaveBeenCalledWith(...) can be satisfied by calls from earlier test cases (false positives / order-dependent). Add a beforeEach(() => mockRunEmbeddedPiAgent.mockClear())/vi.clearAllMocks() or assert against the last/nth call (e.g., toHaveBeenLastCalledWith).

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/hooks/llm-slug-generator.test.ts
Line: 19:22

Comment:
**Mocks not reset**

`mockRunEmbeddedPiAgent` is shared across all tests and its call history is never cleared, so assertions like `toHaveBeenCalledWith(...)` can be satisfied by calls from earlier test cases (false positives / order-dependent). Add a `beforeEach(() => mockRunEmbeddedPiAgent.mockClear())`/`vi.clearAllMocks()` or assert against the last/nth call (e.g., `toHaveBeenLastCalledWith`).

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

Additional Comments (1)

src/hooks/llm-slug-generator.ts
Agent override ignored

This resolves agentId but then uses resolveConfiguredModelRef({ cfg: params.cfg, ... }), which ignores per-agent model overrides (see resolveDefaultModelForAgent in src/agents/model-selection.ts). If an agent has a specific primary model configured, slug generation will still use the global defaults model instead of the agent’s model.

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/hooks/llm-slug-generator.ts
Line: 28:31

Comment:
**Agent override ignored**

This resolves `agentId` but then uses `resolveConfiguredModelRef({ cfg: params.cfg, ... })`, which ignores per-agent model overrides (see `resolveDefaultModelForAgent` in `src/agents/model-selection.ts`). If an agent has a specific primary model configured, slug generation will still use the global defaults model instead of the agent’s model.

How can I resolve this? If you propose a fix, please make it concise.

[openclaw-barnacle]

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.