- Track loops by `(tool, inputHash)` not just tool name (#2)
- Use "Failed after" narrative for error traces (#3)
- Add keyboard accessibility to viewer tabs (role, tabindex, Enter/Space) (#4)
- Use full command as dedup key, not `slice(0,60)` (#5)
- Sort timeline events by time before rendering (#6)
- Pass `tracesDir` to footer text in `listRecaps` (#7)
- Increase `MAX_RECAPS` to 100, add eviction warning log (#8)
- Resolve assistant `parentID` for recap enrichment (#9)
- Remove unused `tracer` variable in test (#10)
- Clarify `--no-trace` backward-compat flag in docs (#1)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…381)

* feat: rename tracer to recap with loop detection, post-session summary, and enhanced viewer

- Rename `Tracer` class to `Recap` with backward-compat aliases
- Rename CLI command `trace` to `recap` (hidden `trace` alias preserved)
- Add loop detection: flags repeated tool calls with same input (3+ in last 10)
- Add post-session summary: `narrative`, `topTools`, `loops` in trace output
- New Summary tab (default) in HTML viewer with:
  - Truncated prompt with expand toggle
  - Files changed with SQL diff previews
  - Tool-agnostic outcome extraction (dbt, pytest, Airflow, pip, SQL)
  - Deduped dbt commands with pass/fail status, clickable to waterfall
  - Smart command grouping (boring ls/cd collapsed, meaningful shown)
  - Error details with resolution tracking
  - Cost breakdown in collapsible section
- Virality: Share Recap (self-contained HTML download), Copy Summary (markdown),
  Copy Link, branded footer
- Fix XSS: timeline items escaped with `e()`
- Fix memory leak: per-session `sessionUserMsgIds` with cleanup on eviction
- Fix JS syntax: onclick quote escaping in collapsible section
- Bound `toolCallHistory` to prevent unbounded growth (cap at 200)
- Summary view wrapped in try-catch for visible error messages
- Update all 13 test files for rename + 8 new adversarial viewer tests
- Update docs: `tracing.md` → `recap.md`, CLI/TUI references updated

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: share/copy buttons scoping bug + `t.text` undefined + adversarial viewer tests

- Fix critical bug: Share Recap and Copy Summary buttons referenced variables
  from Summary IIFE scope — rewrote `buildMarkdownSummary` to be self-contained
- Fix `t.text` → `t.result` in narrative (was rendering "undefined")
- Fix `sessionUserMsgIds` not cleaned on MAX_RECAPS eviction (memory leak)
- Fix zero cost display: show `$0.00` instead of em-dash
- Add try-catch error boundary around Summary view rendering
- Add 8 adversarial viewer tests: XSS, NaN/Infinity, null metadata,
  200+ spans, JS syntax validation, tool-agnostic outcomes, backward compat

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: address all 10 CodeRabbit review comments

- Track loops by `(tool, inputHash)` not just tool name (#2)
- Use "Failed after" narrative for error traces (#3)
- Add keyboard accessibility to viewer tabs (role, tabindex, Enter/Space) (#4)
- Use full command as dedup key, not `slice(0,60)` (#5)
- Sort timeline events by time before rendering (#6)
- Pass `tracesDir` to footer text in `listRecaps` (#7)
- Increase `MAX_RECAPS` to 100, add eviction warning log (#8)
- Resolve assistant `parentID` for recap enrichment (#9)
- Remove unused `tracer` variable in test (#10)
- Clarify `--no-trace` backward-compat flag in docs (#1)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* docs: add screenshots and update recap viewer documentation

- Add Summary tab and full-page screenshots to docs
- Update viewer section with 5-tab description
- Detail what Summary tab shows: files changed, outcomes, timeline, cost
- Add screenshot at top of recap.md for quick visual reference

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* docs: move Recap to Use section, Telemetry to Reference

- Move Recap from Configure > Observability to Use (peer to Commands, Skills)
- Move Telemetry from Configure > Observability to Reference (internal analytics)
- Remove the Observability section entirely

Recap is a feature users interact with after sessions, not a config setting.
Telemetry is internal product analytics, not user-facing observability.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: viewer UX improvements from 100-trace analysis

- Collapse Files Changed after 5 entries with "Show all N files" toggle
- Rename "GENS" → "LLM Calls" in header cards
- Hide Tokens card when cost is $0 (not actionable without cost context)
- Hide Cost metric card when $0.00 (wasted space)
- Add prominent error summary banner right after header metrics
- Improved dbt outcome detection: catch [PASS], [ERROR], N of M, Compilation Error
- Outcome detection rate improved from 18% → 33% across 100 real traces
- Updated doc screenshots with cleaner samples

Tested across 100 real production traces: 0 crashes, 0 JS errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: always show Cost and Tokens cards

$0.00 is a valid cost (Anthropic Max plan). Hiding it implies
we don't support cost tracking.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: tool-agnostic outcome extraction for schema, validation, SQL, lineage tools

500-trace analysis revealed:
- Schema tasks: 0% outcome visibility → 100%
- Validation tasks: 0% outcome visibility → 100%
- SQL tasks: 55% outcome visibility → 100%

Added outcome extraction for:
- schema_inspect, lineage_check, altimate_core_validate results
- SQL error messages (not just row counts)
- Improved empty session display (shows prompt if available)

Tested across 500 diverse synthetic traces (SQL, Airflow, Dagster,
Python, schema, validation, migration, connectors) + 100 real traces.
0 crashes, 0 JS errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: address 4 new CodeRabbit review comments

- Add `inputHash` to `TraceFile.summary.loops` schema type (#11)
- Replace `startTrace()` API name with plain language in docs (#12)
- Use `CSS.escape()` for spanId in querySelector to handle special chars (#13)
- Sort spans by startTime before searching for error resolution (#14)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: round 3 review — sort spans once, clean narrative for 0 LLM calls

- Sort spans once before error resolution loop instead of per-error (perf)
- Narrative omits "Made 0 LLM calls" for tool-only sessions (UX)
- Updated tests to match new narrative format

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: add missing `altimate_change` markers for recap rename in upstream-shared files

Wrap renamed code (Tracer→Recap, trace→recap) with markers so the
Marker Guard CI check passes. The diff-based checker uses -U5 context
windows per hunk — markers must be close enough to added lines to
appear within each hunk's context.

Files fixed:
- `trace.ts` — handler body, option descriptions, viewer message, compat alias
- `app.tsx` — recapViewerServer return, openRecapInBrowser function
- `dialog-trace-list.tsx` — error title, Recaps title, compat alias
- `worker.ts` — getOrCreateRecap, part events, session title/finalization
- `index.ts` — .command(RecapCommand) registration

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: add altimate_change markers to all upstream-shared files

Marker Guard CI was failing — 5 upstream-shared files had custom
code (recap rename) without altimate_change markers.

Fixed: trace.ts, app.tsx, dialog-trace-list.tsx, worker.ts, index.ts

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: type errors in training-import.test.ts from main merge

Pre-existing type issues from main: mock missing `context`/`rule`
fields and readFile return type mismatch.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

…381)

* feat: rename tracer to recap with loop detection, post-session summary, and enhanced viewer

- Rename `Tracer` class to `Recap` with backward-compat aliases
- Rename CLI command `trace` to `recap` (hidden `trace` alias preserved)
- Add loop detection: flags repeated tool calls with same input (3+ in last 10)
- Add post-session summary: `narrative`, `topTools`, `loops` in trace output
- New Summary tab (default) in HTML viewer with:
  - Truncated prompt with expand toggle
  - Files changed with SQL diff previews
  - Tool-agnostic outcome extraction (dbt, pytest, Airflow, pip, SQL)
  - Deduped dbt commands with pass/fail status, clickable to waterfall
  - Smart command grouping (boring ls/cd collapsed, meaningful shown)
  - Error details with resolution tracking
  - Cost breakdown in collapsible section
- Virality: Share Recap (self-contained HTML download), Copy Summary (markdown),
  Copy Link, branded footer
- Fix XSS: timeline items escaped with `e()`
- Fix memory leak: per-session `sessionUserMsgIds` with cleanup on eviction
- Fix JS syntax: onclick quote escaping in collapsible section
- Bound `toolCallHistory` to prevent unbounded growth (cap at 200)
- Summary view wrapped in try-catch for visible error messages
- Update all 13 test files for rename + 8 new adversarial viewer tests
- Update docs: `tracing.md` → `recap.md`, CLI/TUI references updated

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: share/copy buttons scoping bug + `t.text` undefined + adversarial viewer tests

- Fix critical bug: Share Recap and Copy Summary buttons referenced variables
  from Summary IIFE scope — rewrote `buildMarkdownSummary` to be self-contained
- Fix `t.text` → `t.result` in narrative (was rendering "undefined")
- Fix `sessionUserMsgIds` not cleaned on MAX_RECAPS eviction (memory leak)
- Fix zero cost display: show `$0.00` instead of em-dash
- Add try-catch error boundary around Summary view rendering
- Add 8 adversarial viewer tests: XSS, NaN/Infinity, null metadata,
  200+ spans, JS syntax validation, tool-agnostic outcomes, backward compat

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: address all 10 CodeRabbit review comments

- Track loops by `(tool, inputHash)` not just tool name (#2)
- Use "Failed after" narrative for error traces (#3)
- Add keyboard accessibility to viewer tabs (role, tabindex, Enter/Space) (#4)
- Use full command as dedup key, not `slice(0,60)` (#5)
- Sort timeline events by time before rendering (#6)
- Pass `tracesDir` to footer text in `listRecaps` (#7)
- Increase `MAX_RECAPS` to 100, add eviction warning log (#8)
- Resolve assistant `parentID` for recap enrichment (#9)
- Remove unused `tracer` variable in test (#10)
- Clarify `--no-trace` backward-compat flag in docs (#1)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* docs: add screenshots and update recap viewer documentation

- Add Summary tab and full-page screenshots to docs
- Update viewer section with 5-tab description
- Detail what Summary tab shows: files changed, outcomes, timeline, cost
- Add screenshot at top of recap.md for quick visual reference

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* docs: move Recap to Use section, Telemetry to Reference

- Move Recap from Configure > Observability to Use (peer to Commands, Skills)
- Move Telemetry from Configure > Observability to Reference (internal analytics)
- Remove the Observability section entirely

Recap is a feature users interact with after sessions, not a config setting.
Telemetry is internal product analytics, not user-facing observability.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: viewer UX improvements from 100-trace analysis

- Collapse Files Changed after 5 entries with "Show all N files" toggle
- Rename "GENS" → "LLM Calls" in header cards
- Hide Tokens card when cost is $0 (not actionable without cost context)
- Hide Cost metric card when $0.00 (wasted space)
- Add prominent error summary banner right after header metrics
- Improved dbt outcome detection: catch [PASS], [ERROR], N of M, Compilation Error
- Outcome detection rate improved from 18% → 33% across 100 real traces
- Updated doc screenshots with cleaner samples

Tested across 100 real production traces: 0 crashes, 0 JS errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: always show Cost and Tokens cards

$0.00 is a valid cost (Anthropic Max plan). Hiding it implies
we don't support cost tracking.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: tool-agnostic outcome extraction for schema, validation, SQL, lineage tools

500-trace analysis revealed:
- Schema tasks: 0% outcome visibility → 100%
- Validation tasks: 0% outcome visibility → 100%
- SQL tasks: 55% outcome visibility → 100%

Added outcome extraction for:
- schema_inspect, lineage_check, altimate_core_validate results
- SQL error messages (not just row counts)
- Improved empty session display (shows prompt if available)

Tested across 500 diverse synthetic traces (SQL, Airflow, Dagster,
Python, schema, validation, migration, connectors) + 100 real traces.
0 crashes, 0 JS errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: address 4 new CodeRabbit review comments

- Add `inputHash` to `TraceFile.summary.loops` schema type (#11)
- Replace `startTrace()` API name with plain language in docs (#12)
- Use `CSS.escape()` for spanId in querySelector to handle special chars (#13)
- Sort spans by startTime before searching for error resolution (#14)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: round 3 review — sort spans once, clean narrative for 0 LLM calls

- Sort spans once before error resolution loop instead of per-error (perf)
- Narrative omits "Made 0 LLM calls" for tool-only sessions (UX)
- Updated tests to match new narrative format

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: add missing `altimate_change` markers for recap rename in upstream-shared files

Wrap renamed code (Tracer→Recap, trace→recap) with markers so the
Marker Guard CI check passes. The diff-based checker uses -U5 context
windows per hunk — markers must be close enough to added lines to
appear within each hunk's context.

Files fixed:
- `trace.ts` — handler body, option descriptions, viewer message, compat alias
- `app.tsx` — recapViewerServer return, openRecapInBrowser function
- `dialog-trace-list.tsx` — error title, Recaps title, compat alias
- `worker.ts` — getOrCreateRecap, part events, session title/finalization
- `index.ts` — .command(RecapCommand) registration

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: add altimate_change markers to all upstream-shared files

Marker Guard CI was failing — 5 upstream-shared files had custom
code (recap rename) without altimate_change markers.

Fixed: trace.ts, app.tsx, dialog-trace-list.tsx, worker.ts, index.ts

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* fix: type errors in training-import.test.ts from main merge

Pre-existing type issues from main: mock missing `context`/`rule`
fields and readFile return type mismatch.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

Fixes five correctness, reliability, and portability issues surfaced by the
consensus code review of this branch.

CRITICAL #1 — Cross-dialect partitioned diff (`data-diff.ts`):
`runPartitionedDiff` built one partition WHERE clause with `sourceDialect`
and passed it as shared `where_clause` to the recursive `runDataDiff`, which
applied it to both warehouses identically. Cross-dialect partition mode
(MSSQL → Postgres) failed because the target received T-SQL
`DATETRUNC`/`CONVERT(DATE, …, 23)`. Now builds per-side WHERE using each
warehouse's dialect and bakes it into dialect-quoted subquery SQL for source
and target independently. The existing side-aware CTE injection handles the
rest.

MAJOR #2 — Azure AD token caching and refresh (`sqlserver.ts`):
`acquireAzureToken` fetched a fresh token on every `connect()` and embedded
it in the pool config with no refresh. Long-lived sessions silently failed
when the ~1h token expired. Adds a module-scoped cache keyed by
`(resource, client_id)` with proactive refresh 5 min before expiry, parsing
`expiresOnTimestamp` from `@azure/identity` or the JWT `exp` claim from the
`az` CLI fallback. Exposes `_resetTokenCacheForTests` for isolation.

MAJOR #3 — `joindiff` + cross-warehouse guard (`data-diff.ts`):
Explicit `algorithm: "joindiff"` combined with different warehouses produced
broken SQL (one task referencing two CTE aliases with only one injected).
Now returns an early error with a clear message steering users to `hashdiff`
or `auto`. Cross-warehouse detection switched from warehouse-name string
compare to dialect compare, matching the underlying SQL-divergence invariant.

MAJOR #4 — Dialect-aware identifier quoting in CTE wrapping (`data-diff.ts`):
`resolveTableSources` wrapped plain-table names with ANSI double-quotes for
all dialects. T-SQL/Fabric require `QUOTED_IDENTIFIER ON` for this to work;
default for `mssql`/tedious is ON, but user contexts (stored procs, legacy
collations) can override. Now accepts source/target dialect parameters and
delegates to `quoteIdentForDialect`, which was hoisted to module scope so
it can be reused across partition and CTE paths.

MAJOR #5 — Configurable Azure resource URL (`sqlserver.ts`, `normalize.ts`):
Token acquisition hardcoded `https://database.windows.net/`, blocking Azure
Government, Azure China, and sovereign-cloud customers. Now honours an
explicit `azure_resource_url` config field and otherwise infers the URL
from the host suffix (`.usgovcloudapi.net`, `.chinacloudapi.cn`). Adds the
usual camelCase/snake_case aliases in the SQL Server normalizer.

Also surfaces Azure auth error causes: if both `@azure/identity` and `az`
CLI fail, the thrown error includes both hints (redacted) so users know why
rather than seeing the generic "install @azure/identity or run az login"
message.

Tests: adds `data-diff-cross-dialect.test.ts` covering the cross-dialect
partition WHERE routing and the `joindiff` guard; extends
`data-diff-cte.test.ts` with dialect-aware quoting assertions for tsql,
fabric, and mysql; extends `sqlserver-unit.test.ts` with cache hit / expiry
refresh / client-id keyed cache tests, commercial/gov/china/custom resource
URL resolution, and the combined-error-hints surface.

All 41 sqlserver driver tests, 24 data-diff orchestrator tests, and 214
normalize/connections tests pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* fix: use synchronous DuckDB constructor to avoid bun runtime timeout

Bun's runtime never fires native addon async callbacks, so the async
`new duckdb.Database(path, opts, callback)` form would hit the 2-second
timeout fallback on every connection attempt.

Switch to the synchronous constructor form `new duckdb.Database(path)` /
`new duckdb.Database(path, opts)` which throws on error and completes
immediately in both Node and bun runtimes.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

* revert: restore async DuckDB constructor — sync change was bogus

The async callback form with 2s fallback was already working correctly
at e3df5a4. The timeout was caused by a missing duckdb .node binary,
not a bun incompatibility.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

* feat: add MSSQL/Fabric dialect mapping and data-parity support

- Add `warehouseTypeToDialect()` mapping: sqlserver→tsql, mssql→tsql,
  fabric→fabric, postgresql→postgres, mariadb→mysql. Fixes critical
  serde mismatch where Rust engine rejects raw warehouse type names.
- Update both `resolveDialect()` functions to use the mapping
- Add MSSQL/Fabric cases to `dateTruncExpr()` — DATETRUNC(DAY, col)
- Add locale-safe date literal casting via CONVERT(DATE, ..., 23)
- Register `fabric` in DRIVER_MAP (reuses sqlserver TDS driver)
- Add `fabric` normalize aliases in normalize.ts
- Add 15 SQL Server driver unit tests (TOP injection, truncation,
  schema introspection, connection lifecycle, result format)
- Add 9 dialect mapping unit tests

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* feat: add Azure AD authentication to SQL Server driver (7 flows)

- Support all 7 Azure AD / Entra ID auth types in `sqlserver.ts`:
  `azure-active-directory-password`, `access-token`, `service-principal-secret`,
  `msi-vm`, `msi-app-service`, `azure-active-directory-default`, `token-credential`
- Force TLS encryption for all Azure AD connections
- Dynamic import of `@azure/identity` for `DefaultAzureCredential`
- Add normalize aliases for Azure AD config fields (`authentication`,
  `azure_tenant_id`, `azure_client_id`, `azure_client_secret`, `access_token`)
- Add `fabric: SQLSERVER_ALIASES` to DRIVER_ALIASES
- Add 10 Azure AD unit tests covering all auth flows, encryption,
  and `DefaultAzureCredential` with managed identity

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* docs: add MSSQL and Microsoft Fabric documentation to data-parity SKILL.md

- Add SQL Server / Fabric schema inspection query in Step 2
- Add "SQL Server and Microsoft Fabric" section with:
  - Supported configurations table (sqlserver, mssql, fabric)
  - Fabric connection guide with Azure AD auth types
  - Algorithm behavior notes (joindiff vs hashdiff selection)

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: delegate Azure AD credential creation to tedious and remove underscore column filter

- **Azure AD auth**: Pass `azure-active-directory-*` types directly to tedious
  instead of constructing `DefaultAzureCredential` ourselves. Tedious imports
  `@azure/identity` internally and creates credentials — avoids bun CJS/ESM
  `isTokenCredential` boundary issue that caused "not an instance of the token
  credential class" errors.
- **Auth shorthands**: Map `CLI`, `default`, `password`, `service-principal`,
  `msi`, `managed-identity` to their full tedious type names.
- **Column filter**: Remove `_.startsWith("_")` filter from `execute()` result
  columns — it stripped legitimate aliases like `_p` used by partition discovery,
  causing partitioned diffs to return empty results.
- **Tests**: Remove `@azure/identity` mock (no longer imported by driver),
  update auth assertions, add shorthand mapping tests, fix column filter test.
- **Verified**: All 97 driver tests pass. Full data-diff pipeline tested against
  real MSSQL server (profile, joindiff, auto, where_clause, partitioned).

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: upgrade `mssql` to v12 with `ConnectionPool` isolation and row flattening

- Upgrade `mssql` from v11 to v12 (`tedious` 18 → 19)
- Use explicit `ConnectionPool` instead of global `mssql.connect()` to
  isolate multiple simultaneous connections
- Flatten unnamed column arrays — `mssql` merges unnamed columns (e.g.
  `SELECT COUNT(*), SUM(...)`) into a single array under the empty-string
  key; restore positional column values
- Proper column name resolution: compare `namedKeys.length` against
  flattened row length, fall back to synthetic `col_0`, `col_1`, etc.
- Update test mock to export `ConnectionPool` class and `createMockPool`

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: resolve TypeScript spread-type errors in Azure AD conditional options

Use ternary expressions (`x ? {...} : {}`) instead of short-circuit
(`x && {...}`) to avoid spreading a boolean value.

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: resolve cubic review findings on MSSQL/Fabric PR

- P1: restrict `flattenRow` to only spread the empty-string key (`""`)
  where mssql merges unnamed columns, preserving legitimate array values
- P2: escape single quotes in `partitionValue` for date-mode branches in
  `buildPartitionWhereClause` (categorical mode already escaped)
- P2: add `fabric` to `PASSWORD_DRIVERS` set in registry for consistent
  password validation alongside `sqlserver`/`mssql`
- P2: fallback to `"(no values)"` when `d.values` is nullish to prevent
  template literal coercing `undefined` to the string `"undefined"`

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* test: add fabric connection path and flattenRow coverage

- sqlserver-unit: 3 tests for unnamed column flattening — verifies only
  the empty-string key is spread, legitimate named arrays are preserved
- driver-normalize: fabric type uses SQLSERVER_ALIASES (server → host,
  trustServerCertificate → trust_server_certificate)
- connections: fabric type is recognized in DRIVER_MAP and listed correctly

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* docs: document minimum versions and make @azure/identity optional

- Add "Minimum Version Requirements" table to SKILL.md covering SQL Server
  2022+, mssql v12, and @azure/identity v4 with rationale for each
- Document auth shorthands (CLI, default, password, service-principal, msi)
- Move @azure/identity from dependencies to optional peerDependencies so
  it is NOT installed by default — only required for Azure AD auth
- Add runtime check in sqlserver driver: if Azure AD auth type is requested
  but @azure/identity is missing, throw a clear install instruction error

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: acquire Azure AD tokens directly to bypass Bun browser-bundle resolution

- For `azure-active-directory-default` (CLI/default auth), acquire token
  ourselves instead of delegating to tedious's internal `@azure/identity`
- Strategy: try `DefaultAzureCredential` first, fall back to `az` CLI subprocess
- Bypasses Bun resolving `@azure/identity` to browser bundle where
  `DefaultAzureCredential` is a non-functional stub
- Also bypasses CJS/ESM `isTokenCredential` boundary mismatch
- All 31 driver unit tests pass, verified against real Fabric endpoint

Co-Authored-By: Claude Opus 4.6 <[email protected]>

* fix: auto-acquire Azure AD token for `azure-active-directory-access-token` when none supplied

The `azure-active-directory-access-token` branch passed
`token: config.token ?? config.access_token` to tedious. When neither
field was set on a connection (e.g. a `fabric-migration` entry that
declared the auth type but no token), tedious threw:

    TypeError: The "config.authentication.options.token" property
    must be of type string

This blocked any Fabric/MSSQL config that relied on ambient credentials
(Azure CLI / managed identity) but used the explicit
`azure-active-directory-access-token` type instead of the `default`
shorthand.

Refactor token acquisition (`DefaultAzureCredential` → `az` CLI
fallback) into a shared `acquireAzureToken()` helper used by both the
`default` path and the `access-token` path when no token was supplied.
Callers that pass an explicit token are unchanged.

Also harden `mock.module("node:child_process", ...)` in
`sqlserver-unit.test.ts` to spread the real module so sibling tests in
the same `bun test` run keep access to `spawn` / `exec` / `fork`.

Tests: 110 pass, 0 fail in `packages/drivers`.

Co-Authored-By: Claude Opus 4.7 <[email protected]>

* fix: side-aware CTE injection for cross-warehouse `data_diff` SQL-query mode

When `source` and `target` are both SQL queries, `resolveTableSources` wraps
them as `__diff_source` / `__diff_target` CTEs and the executor prepends the
combined `WITH …` block to every engine-emitted task. T-SQL and Fabric
parse-bind every CTE body even when unreferenced, so a task routed to the
source warehouse failed to resolve the target-only base table referenced
inside the unused `__diff_target` CTE (and vice versa), producing
`Invalid object name` errors from the wrong warehouse.

Return side-specific prefixes from `resolveTableSources` alongside the
combined one, and have the executor loop in `runDataDiff` pick the source
or target prefix per task when `source_warehouse !== target_warehouse`.
Same-warehouse behaviour is unchanged.

Adds `data-diff-cte.test.ts` covering plain-name passthrough, both-query
wrapping, side-specific CTE isolation, and CTE merging with engine-emitted
`WITH` clauses (10 tests).

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* chore: regenerate `bun.lock` to match drivers `peerDependencies` layout

Commit 333a45c moved `@azure/identity` from `optionalDependencies` to
`peerDependencies` with `optional: true` in `packages/drivers/package.json`,
but the lockfile was not regenerated. That left CI under `--frozen-lockfile`
broken and made fresh installs silently diverge from the committed state.

Running `bun install` brings the lockfile in sync: `@azure/identity` is
recorded as an optional peer, and its transitive pins (`@azure/msal-browser`,
`@azure/msal-common`, `@azure/msal-node`) re-resolve to the versions required
by `tedious` and `snowflake-sdk`, matching the reachable runtime surface.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* fix: address all CRITICAL/MAJOR findings from multi-model review

Fixes five correctness, reliability, and portability issues surfaced by the
consensus code review of this branch.

CRITICAL #1 — Cross-dialect partitioned diff (`data-diff.ts`):
`runPartitionedDiff` built one partition WHERE clause with `sourceDialect`
and passed it as shared `where_clause` to the recursive `runDataDiff`, which
applied it to both warehouses identically. Cross-dialect partition mode
(MSSQL → Postgres) failed because the target received T-SQL
`DATETRUNC`/`CONVERT(DATE, …, 23)`. Now builds per-side WHERE using each
warehouse's dialect and bakes it into dialect-quoted subquery SQL for source
and target independently. The existing side-aware CTE injection handles the
rest.

MAJOR #2 — Azure AD token caching and refresh (`sqlserver.ts`):
`acquireAzureToken` fetched a fresh token on every `connect()` and embedded
it in the pool config with no refresh. Long-lived sessions silently failed
when the ~1h token expired. Adds a module-scoped cache keyed by
`(resource, client_id)` with proactive refresh 5 min before expiry, parsing
`expiresOnTimestamp` from `@azure/identity` or the JWT `exp` claim from the
`az` CLI fallback. Exposes `_resetTokenCacheForTests` for isolation.

MAJOR #3 — `joindiff` + cross-warehouse guard (`data-diff.ts`):
Explicit `algorithm: "joindiff"` combined with different warehouses produced
broken SQL (one task referencing two CTE aliases with only one injected).
Now returns an early error with a clear message steering users to `hashdiff`
or `auto`. Cross-warehouse detection switched from warehouse-name string
compare to dialect compare, matching the underlying SQL-divergence invariant.

MAJOR #4 — Dialect-aware identifier quoting in CTE wrapping (`data-diff.ts`):
`resolveTableSources` wrapped plain-table names with ANSI double-quotes for
all dialects. T-SQL/Fabric require `QUOTED_IDENTIFIER ON` for this to work;
default for `mssql`/tedious is ON, but user contexts (stored procs, legacy
collations) can override. Now accepts source/target dialect parameters and
delegates to `quoteIdentForDialect`, which was hoisted to module scope so
it can be reused across partition and CTE paths.

MAJOR #5 — Configurable Azure resource URL (`sqlserver.ts`, `normalize.ts`):
Token acquisition hardcoded `https://database.windows.net/`, blocking Azure
Government, Azure China, and sovereign-cloud customers. Now honours an
explicit `azure_resource_url` config field and otherwise infers the URL
from the host suffix (`.usgovcloudapi.net`, `.chinacloudapi.cn`). Adds the
usual camelCase/snake_case aliases in the SQL Server normalizer.

Also surfaces Azure auth error causes: if both `@azure/identity` and `az`
CLI fail, the thrown error includes both hints (redacted) so users know why
rather than seeing the generic "install @azure/identity or run az login"
message.

Tests: adds `data-diff-cross-dialect.test.ts` covering the cross-dialect
partition WHERE routing and the `joindiff` guard; extends
`data-diff-cte.test.ts` with dialect-aware quoting assertions for tsql,
fabric, and mysql; extends `sqlserver-unit.test.ts` with cache hit / expiry
refresh / client-id keyed cache tests, commercial/gov/china/custom resource
URL resolution, and the combined-error-hints surface.

All 41 sqlserver driver tests, 24 data-diff orchestrator tests, and 214
normalize/connections tests pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* fix: address PR #705 bot review findings (coderabbitai + cubic + copilot)

Addresses the remaining issues raised by coderabbitai, cubic-dev-ai, and the
Copilot PR reviewer on top of the multi-model consensus fix.

### CRITICAL

- **`@azure/identity` peer dep removed** (`drivers/package.json`)
  `mssql@12` → `tedious@19` bundles `@azure/identity ^4.2.1` as a regular
  dependency. Declaring it here as an optional peer was redundant and caused
  transitive-version-drift concerns. Users get the correct version
  automatically through the tedious chain; our lazy import handles the
  browser-bundle edge case itself.

### MAJOR

- **Cross-dialect date partition literal normalization** (`data-diff.ts`)
  `buildPartitionDiscoverySQL` on MSSQL returns a JS `Date` object, stringified
  upstream as `"Mon Jan 01 2024 …"`. `CONVERT(DATE, …, 23)` rejects that format.
  Normalize `partitionValue` to ISO `yyyy-mm-dd` before dialect casting so the
  T-SQL/Fabric path works end-to-end on dates discovered from MSSQL sources.

- **`crossWarehouse` uses resolved warehouse identity** (`data-diff.ts`)
  Previous commit gated on dialect compare, which treated two independent
  MSSQL instances as "same warehouse" and would have let `joindiff` route a
  JOIN through a warehouse that can't resolve the other side's base tables.
  Now resolves both sides' warehouse name (falling back to the default
  warehouse when a side is omitted) and compares identities — identity-based
  gating handles both the "undefined vs default" case (cubic) and the
  "same-dialect, different instance" case (Copilot).

- **Drop `mssql.connect()` fallback** (`sqlserver.ts`)
  `mssql@^12` guarantees `ConnectionPool` as a named export. The fallback
  silently re-introduced the global-shared-pool bug this branch was added
  to fix. Now throws a descriptive error if `ConnectionPool` is missing —
  cross-database pool interference cannot regress.

- **Non-string `config.authentication` guarded** (`sqlserver.ts`)
  Caller passing a pre-built `{ type, options }` block (or `null`) previously
  crashed with `TypeError: rawAuth.toLowerCase is not a function`. Now only
  applies the shorthand lookup when `rawAuth` is a string; other values pass
  through so tedious can handle them or reject them with its own error.

- **Unknown `azure-active-directory-*` subtype fails fast** (`sqlserver.ts`)
  Typos or future tedious subtypes previously dropped through all `else if`
  branches, producing a config with `encrypt: true` but no `authentication`
  block. tedious then surfaced an opaque error far from the root cause.
  Now throws with the offending subtype and the supported list.

- **`execSync` replaced with async `exec`** (`sqlserver.ts`)
  The `az account get-access-token` CLI fallback previously blocked the
  event loop for up to 15s. Switched to `util.promisify(exec)` so the
  connection path stays non-blocking.

- **Mixed named + unnamed column derivation preserves headers** (`sqlserver.ts`)
  Previously `SELECT name, COUNT(*), SUM(x)` produced either `["name", ""]`
  (blank header) or `["col_0", "col_1", "col_2"]` (lost `name`). Rewrote
  column/row derivation to iterate in one pass, preserving known named
  columns and synthesizing `col_N` only for expanded `""`-key positions.

### MINOR

- **`(no values)` fallback for empty `diff_row.values` array** (`tools/data-diff.ts`)
  `[].join(" | ") ?? "(no values)"` never fires because `""` is falsy-but-not-
  nullish. Gate on `d.values?.length` instead.

### Test / docs

- `sqlserver-unit.test.ts`: token-cache client-id test now counts actual
  `getToken` invocations (previous version only verified both got the same
  mocked token, which proved nothing about keying).
- `sqlserver-unit.test.ts`: "empty result" test now mirrors the real mssql
  shape (`recordset.columns` is a property *on* the recordset array, not a
  sibling key).
- `sqlserver-unit.test.ts`: added mixed-column regression tests — "name +
  COUNT + SUM" and "single unnamed column" — to lock in the derivation fix.
- `sqlserver-unit.test.ts`: stubbed async `exec` via `util.promisify.custom`
  so tests drive both the `execSync` legacy path and the new async path.
- `SKILL.md`: Fabric config fenced block now declares `yaml` (markdownlint
  MD040).

All tests: 43/43 sqlserver driver + 238/238 opencode test suite.

Attribution: findings identified by coderabbitai, cubic-dev-ai, and the
Copilot PR reviewer.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* chore: drop stale `@azure/identity` peer-dep entries from `bun.lock`

Commit 38cfb0e removed `@azure/identity` from the drivers package's
`peerDependencies` (tedious already bundles it), but the lockfile's
`packages/drivers` workspace section still carried the corresponding
`peerDependencies` and `optionalPeers` blocks. CI running
`bun install --frozen-lockfile` would fail on the drift.

Minimal edit — just removes the two stale blocks. No resolution changes
(`bun install --frozen-lockfile` passes with "no changes").

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* fix: CI — isolate `data-diff-cross-dialect` tests from other files

The prior integration-style test mocked the Registry module globally with
`mock.module(".../registry", ...)`, which leaks across all test files in
bun:test's single-process runner. That caused 14 unrelated tests in
`connections.test.ts`, `telemetry-safety.test.ts`, and
`dbt-first-execution.test.ts` to fail in CI.

Additionally, the test relied on `mock.module("@altimateai/altimate-core")`
to supply a fake `DataParitySession`. The npm-published 0.2.6 of that
package does not export `DataParitySession` (sessions are only in the
locally-built `altimate-core-internal` binary), and Bun's `mock.module`
cannot override a package that another test file has already imported —
so the integration test was structurally unreliable.

Resolution:

1. **Export pure SQL-builder helpers** from `data-diff.ts`
   (`dateTruncExpr`, `buildPartitionWhereClause`) and unit-test them
   directly. No module mocking required; the test directly exercises the
   logic the CRITICAL/MAJOR fix changed.

2. **Move the `joindiff` + cross-warehouse guard earlier** in `runDataDiff`
   — before the NAPI import. Semantically identical for callers (guard
   still fires, same error message, `steps: 0`), but now it can be
   integration-tested without any NAPI mock. Preserves end-to-end
   wiring coverage for the guard.

3. **Rewrite `data-diff-cross-dialect.test.ts`** as pure-function unit
   tests for the partition WHERE logic + a real `runDataDiff` call for
   the joindiff guard. No more cross-file mock pollution.

Functionality unchanged:
- `runDataDiff` behavior for real callers is identical. The only
  observable difference is error-ordering: if a caller simultaneously
  omits NAPI and passes `joindiff + cross-warehouse`, they now get the
  "joindiff requires same warehouse" error instead of the NAPI-missing
  error. That's strictly better UX — NAPI availability is a deployment
  concern, `joindiff`+cross-warehouse is a user error.
- `buildPartitionWhereClause` and `dateTruncExpr` are now exported but
  semantically unchanged — same inputs, same outputs.

Test results:
- 2821 altimate tests pass, 0 fail
- 43 sqlserver driver tests pass, 0 fail
- The 19 remaining full-suite failures (`mcp/`, `tool/project-scan`,
  `plan-approval-phrase`) are pre-existing on `main` and unrelated.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

* fix: follow-up PR bot review findings (cubic P1/P2 + coderabbit MAJOR/MINOR)

Addresses 5 substantive issues raised by the latest round of bot reviews.

### P1 / MAJOR

- **MySQL/MariaDB week-partition values no longer corrupted** (cubic P1,
  data-diff.ts:610) — the prior ISO `yyyy-mm-dd` normalization applied to
  every dialect silently rewrote MySQL `DATE_FORMAT(%Y-%u)` outputs like
  `"2024-42"` into invalid dates, producing WHERE clauses that never
  match. Scope the normalization to T-SQL / Fabric only — those use
  `CONVERT(DATE, …, 23)` which is the only code path that requires ISO.
  Postgres, MySQL, ClickHouse, BigQuery, Oracle all get the raw value
  verbatim, matching their own `DATE_TRUNC`/`toStartOf*` output.

- **Partitioned diff no longer drops extra_columns** (coderabbit MAJOR,
  data-diff.ts:824) — the partition fix wraps each side as a SELECT
  subquery before recursing. `discoverExtraColumns` skips SQL queries
  (only inspects plain table names), so the recursive `runDataDiff` fell
  through to key-only comparison, silently losing value-level diffs.
  Now `runPartitionedDiff` runs discovery ONCE on the plain source
  table up-front and passes the resolved `extra_columns` explicitly to
  each recursive call. Audit-column exclusion metadata is also
  propagated to the aggregated result for user reporting.

### P2 / MINOR

- **`azure_resource_url` trailing slash normalized** (cubic P2,
  sqlserver.ts:50) — an explicit `"https://custom-host"` (no slash)
  would produce an invalid OAuth scope `"https://custom-host.default"`.
  Enforce a trailing slash in `resolveAzureResourceUrl`.

- **`az account get-access-token` uses `execFile`** (coderabbit,
  sqlserver.ts:200) — replaces `exec(<shell command string>)` with
  `execFile("az", [args])` so user-supplied `azure_resource_url` can't
  introduce shell metacharacters into the command string. Also updates
  the test harness to stub both `exec` and `execFile`.

### Test isolation / coverage

- **Added same-dialect cross-warehouse joindiff test** (cubic,
  data-diff-cross-dialect.test.ts:97) — two MSSQL servers with
  different hosts must still be gated by the joindiff guard; previous
  tests only exercised mixed dialects.

- **Added MySQL week-partition regression tests** — prevent future
  revivals of the dialect-unaware ISO rewrite.

- **Added trailing-slash `azure_resource_url` test.**

Test results:
- 44/44 sqlserver driver tests pass
- 2824/2824 altimate tests pass, 0 fail
- Remaining full-suite failures (`mcp/`, `tool/project-scan`,
  `plan-approval-phrase`) are pre-existing on `main`.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

---------

Co-authored-by: Claude Sonnet 4.6 <[email protected]>