Deep research zero-query path raises NameError before fallback state is initialized

Summary

DeepResearchSkill.deep_research() has a zero-query guard immediately after generate_search_queries(), but the guard returns variables that are initialized only after the guard.

On current main, the control flow is effectively:

serp_queries = await self.generate_search_queries(query, num_queries=breadth)

if not serp_queries:
    return {
        "learnings": all_learnings,
        "visited_urls": all_visited_urls,
        "citations": all_citations,
        "context": all_context,
        "sources": all_sources,
    }

all_learnings = learnings.copy()
all_citations = citations.copy()
all_visited_urls = visited_urls.copy()
all_context = []
all_sources = []

Therefore, if query generation legitimately returns [], the intended graceful-stop branch instead raises NameError because all_learnings (and the other all_* values) do not yet exist.

Why this matters

The function explicitly tries to make the zero-query case a safe termination path:

logger.warning("Deep research generated zero search queries; stopping descent.")

That condition can occur if the model returns malformed/empty output and parse_search_queries_response() yields no valid queries. In that situation, deep research should preserve the accumulated state and stop cleanly rather than failing in the fallback itself.

Expected behavior

If generate_search_queries() returns no queries, deep_research() should return the current accumulated state:

• existing learnings;
• existing citations;
• existing visited_urls;
• empty context/sources for the current level.

No exception should be raised.

Actual behavior

The zero-query branch references these names before assignment:

• all_learnings
• all_visited_urls
• all_citations
• all_context
• all_sources

The first reference raises NameError.

Suggested fix

Initialize the accumulated-state variables before generating/checking the query list:

all_learnings = learnings.copy()
all_citations = citations.copy()
all_visited_urls = visited_urls.copy()
all_context = []
all_sources = []

serp_queries = await self.generate_search_queries(query, num_queries=breadth)

if not serp_queries:
    logger.warning("Deep research generated zero search queries; stopping descent.")
    return {
        "learnings": all_learnings,
        "visited_urls": all_visited_urls,
        "citations": all_citations,
        "context": all_context,
        "sources": all_sources,
    }

Regression coverage

A focused async unit test can stub generate_search_queries() to return [] and verify that:

1. no exception is raised;
2. pre-existing learnings/citations/visited URLs are preserved;
3. current-level context and sources are empty;
4. no nested researcher or retrieval work is started.

I searched the current issue set for this specific zero-query / pre-initialization failure and did not find an existing report for the same root cause.

AI-assisted review disclosure: I used an AI coding assistant to help inspect the control flow and draft this report; the reported behavior is based on the current source structure and is intentionally scoped to the pre-initialization bug above.