lsm / neokai / 24755715908

82%

push

github

fix(db-query): expose space sessions and sdk_messages in space scope (#1581)

* fix(db-query): expose space sessions and sdk_messages in space scope

Extends the db-query MCP tool so that task agents in a Space can query
their own session data, which is needed for operational debugging (e.g.
checking if a research agent's conversation history still exists).

Changes:
- Add ScopeLikeConfig interface + scopeLike field to ScopeTableConfig for
  LIKE-based scope filtering (session ID prefix matching).
- Extend ScopeJoinConfig with optional likePrefix/likeSuffix for LIKE-
  based indirect scope filters.
- Add sessions, sdk_messages, session_groups, and session_group_members to
  SPACE_SCOPE_TABLES with space-scoped filtering:
    sessions          → WHERE id LIKE 'space:<id>:%'
    sdk_messages      → WHERE session_id LIKE 'space:<id>:%'
    session_groups    → WHERE id IN (SELECT group_id FROM
                          session_group_members WHERE session_id LIKE ?)
    session_group_members → WHERE session_id LIKE 'space:<id>:%'
- Remove sdk_messages, session_groups, session_group_members from
  EXCLUDED_TABLE_NAMES (they are now in space scope with filtering).
- Update all three filter-building paths in tools.ts (subquery wrapper,
  aggregate/DISTINCT direct injection) to handle scopeLike and LIKE joins.
- Existing sensitive column blacklist for sessions (config, session_context)
  remains in place for space scope access.
- sessions is intentionally accessible in both global scope (unfiltered,
  for Neo agent) and space scope (prefix-filtered, for space agents).

Tests: update scope-config.test.ts and add 15 new integration tests in
db-query-integration.test.ts covering filtering, isolation, blacklisting,
aggregates, and db_list/describe_table for the new tables.

* docs(db-query): comment why LIKE scopeValue escaping is not needed

The scopeValue fed into LIKE patterns is always a server-side
crypto.randomUUID() string set in the MCP server closure befo... (continued)

9071 of 12358 branches covered (73.4%)

Branch coverage included in aggregate %.

76 of 80 new or added lines in 2 files covered. (95.0%)

66845 of 75678 relevant lines covered (88.33%)

301.18 hits per line