|
1 | | -fix: defensive null guards in tool formatters and DuckDB concurrent access retry (#570) |
2 | | - |
3 | | -- Add null/undefined guards across 8 tool formatters to prevent literal |
4 | | - `undefined` in user-facing output (sql-analyze, schema-inspect, |
5 | | - sql-translate, dbt-manifest, finops-analyze-credits, warehouse-list, |
6 | | - altimate-core-check, altimate-core-rewrite) |
7 | | -- Add `error: msg` to catch block metadata in schema-inspect, |
8 | | - dbt-manifest, warehouse-list so telemetry can classify exceptions |
9 | | -- DuckDB driver: auto-retry in `READ_ONLY` mode on `database is locked` |
10 | | - errors, with clear actionable error message |
11 | | -- Add simulation suite (839 mock + 346 real E2E scenarios) covering |
12 | | - 10 personas x 11 dialects x 14 use-case categories |
| 1 | +feat: add task intent classification telemetry event |
| 2 | + |
| 3 | +Add `task_classified` event emitted at session start with keyword/regex |
| 4 | +classification of the first user message. Categories: debug_dbt, write_sql, |
| 5 | +optimize_query, build_model, analyze_lineage, explore_schema, migrate_sql, |
| 6 | +manage_warehouse, finops, general. |
| 7 | + |
| 8 | +- `classifyTaskIntent()` — pure regex matcher, zero LLM cost, <1ms |
| 9 | +- Includes warehouse type from fingerprint cache |
| 10 | +- Strong/weak confidence levels (1.0 vs 0.5) |
| 11 | +- 15 unit tests covering all intent categories + edge cases |
| 12 | + |
| 13 | +Closes AI-6029 |
13 | 14 |
|
14 | 15 | Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
0 commit comments