[phernandez]

Summary

Stacked on #1964 (refactor/remove-cloud-only-legacy-tools), which removes read_note, the last core caller of search_notes. Merge #1964 first.

Two commits:

1. test(mcp): port search tests to bm_search. Ports the search tests from the retired search_notes function to the published bm_search tool.
2. refactor(mcp): remove search_notes. bm_search is now the only caller, so search.py keeps only the shared query pipeline (about 1,100 src lines deleted).
   • Deleted search_notes, _format_search_error_response, _format_service_unavailable_response, _format_search_markdown, _NO_SEARCH_CRITERIA_MESSAGE, METADATA_FILTERS_DESCRIPTION, and the tools/__init__ re-export.
   • _search_all_projects no longer takes output_format. It always returns the merged rows. MultiProjectRan carries payload, and MultiProjectFailed carries only reason. This removes bm_search's # pragma: no cover text branch.
   • Dropped the status parameter of _build_search_query and the tags parameter of split_tag_shorthand; only search_notes passed them. split_tag_shorthand now returns tags in query order instead of set order.
   • Deleted utils.strict_search_tags (the tags coercer for search_notes) and its tests in tests/test_coerce.py.
   • Ported the remaining callers: test_index_readiness.py, which refactor(mcp): port context and recent-activity tests to bm_context and bm_project #1962 also ports; the parity test in test_tool_posix.py, which refactor(cli)!: rebuild cat/grep/ls/find on bm_* tools and remove tail #1960 deletes; and the search_notes entries in test_tool_contracts.py and implementation_mcp.py.
   • Added tests for bm_search paths that were covered only incidentally before:
     • the explicit-scope refusal;
     • the (more matches from line N) hint;
     • the semantic excerpt in text output;
     • non-Markdown hits dropped by lines;
     • invalid and duplicate project refs;
     • workspace permalink qualification.

Files

Before	After
tests/mcp/test_tool_search.py	tests/mcp/test_tool_bm_search_query.py (new)
tests/mcp/test_mcp_note_type_normalization.py	folded into test_tool_bm_search_query.py
tests/mcp/test_search_total_exactness.py	folded into test_tool_bm_search_query.py
tests/mcp/test_tool_search_temporal.py	tests/mcp/test_tool_bm_search_temporal.py
tests/mcp/tools/test_search_notes_multi_project.py	tests/mcp/tools/test_bm_search_multi_project.py
tests/mcp/tools/test_search_notes_multi_project_temporal.py	folded into test_bm_search_multi_project.py
tests/mcp/tools/test_cjk_search_guidance.py	folded into test_bm_search_multi_project.py
tests/api/v2/test_cjk_search_guidance.py	its MCP assertions now call bm_search
tests/mcp/test_tool_bm_search.py	two tests that replaced _search_all_projects now run the real path or moved to the multi-project file

Ported to real code paths

These tests used to stub get_project_client, resolve_project_and_path and SearchClient and inspect the captured payload. They now do one of two things:

• Run bm_search end to end over notes written with bm_write: title and permalink modes, note types including snake_case canonicalization, items, categories, since, the note_type filter alias, tag shorthand, total exactness per mode, default hybrid mode, Boolean queries defaulting to text, configured default_search_type, never-indexed versus indexed misses, and comma or JSON list strings through the MCP layer.
• Call the shared pure helpers that bm_search uses on every request: _build_search_query, split_tag_shorthand and _search_type_for.

The valid-time doc tests now check bm-search(3) instead of the search_notes docstring.

Remaining monkeypatches, and why

• Multi-project tests: several databases only exist in Cloud. These tests stub the per-database boundary: project discovery, the per-workspace get_client, ScopedSearchClient and the readiness probe. The merge, paging, attribution and failure handling run for real. A new test, test_hits_from_two_local_projects_name_their_project, runs the whole path with no stubs across two real local projects.
• "Filter ran nowhere" temporal test: stubs only ScopedSearchClient, to stand in for a server older than SPEC-82.
• Keyword-only host: two tests patch semantic_runtime_available, because sqlite-vec loads in the test environment.
• Default-mode tests: these install the MCP container the way the server lifespan does. An earlier in-process MCP session leaves its container behind with a stale config, so each test sets its own.

Dropped, with reasons

• Error formatting (TestSearchErrorFormatting, TestSearchToolErrorHandling, invalid search_type returned as a string): bm_search raises tool errors instead of returning guidance markdown. Its refusals are covered in test_tool_bm_search.py.
• _format_search_markdown tests, and the valid-time label in markdown output: this is the legacy text format. bm_search renders with _format_results.
• Workspace memory:// routing through search_notes: bm_search routes memory:// URLs through open_project_target. That path is covered by test_note_targets.py and the project-qualified memory URL test.
• search_notes parameter-annotation tests and tags= tests (comma-split, invalid types, direct-call CLI regression): bm_search has no tags or status parameters, and no CLI calls search_notes any more. The equivalent bm_search list parameters are tested through MCP.
• Multi-project opt-in: bm_search requires an explicit scope, and that refusal is already tested.
• Redundant cases: plain text search, the project-prefixed memory URL and page_size=1 pagination are already covered in test_tool_bm_search.py.

Coverage

search.py and ranked_search.py are both at 100%, measured over the touched test files.

Verification

• just fast-check
• just test-unit-sqlite -k "search or test_index_readiness or test_tool_contracts or test_coerce or ...": 1146 passed. One failure, test_tokenizing_scales_linearly_with_query_length, is a timing test; it passes when rerun on its own.
• just test-unit-postgres -k "bm_search or test_cjk_search_guidance or test_index_readiness or test_tool_contracts or test_coerce or bm_search_parity": all pass, after making one new test backend-neutral.
• just test-int-sqlite -k search: 64 passed

🤖 Generated with Claude Code

https://claude.ai/code/session_01CMYVL1x91B8kJxy7nn2GU8

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T20:22:57.398325Z	dbd4f12	Manual request

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.

[phernandez]

@codex review

[phernandez]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Swish!

Reviewed commit: 1e1f3d9513

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[phernandez]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3c96a93d9f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[phernandez]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: dbd4f120e6

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".