Adds the 2026-05-14 design spec for exposing four new MCP tools
(discover_data, entity_details, dictionary_search, sql_execution),
shipping a ktx-research skill, and introducing an HTTP-only ktx mcp
daemon so external agents can use KTX as a research-capable context
layer.
Annotates the Zod config schema with .describe() text on every field and
adds generateKtxProjectConfigJsonSchema() plus a ktx dev schema command
that prints (or writes) a draft-07 JSON Schema for editors and LLM agents.
* feat(cli): extend `ktx connection test` to every supported driver
Dispatch by driver: native DBs now call `connector.testConnection()`
(was `introspect(dryRun)`), looker/notion/metabase hit their auth
endpoints, and dbt/metricflow/lookml run `git ls-remote` via the
existing `testRepoConnection` helper. Unknown drivers exit 1 with a
listing of supported ones.
* feat(cli): add `ktx connection test --all` summary list
Tests every configured connection in parallel and renders a single
Clack-style list (◇/│/◆/└, green ✓ / red ✗) consistent with sl list,
with per-row detail and a passed/failed footer. Exits non-zero if any
connection fails. Single-id `ktx connection test` output is preserved.
* fix(cli): read metabase status url from api_url
`ktx status` was probing `url` / `base_url` on metabase connections, but
ktx.yaml stores it as `api_url`, so the field always reported "url not
set". Read `api_url` directly and align the warning text with the actual
key.
* refactor(context): validate ktx.yaml with Zod and surface issues in status
- Replace hand-rolled ktx.yaml parsing with a strict Zod schema and
derive KtxProjectConfig types from it.
- Add validateKtxProjectConfig returning structured KtxConfigIssue[]
with migration hints for deprecated keys (ingest.llm,
scan.enrichment.backend, etc.).
- Wire ktx status/doctor to run validation, render schema issues in
plain and JSON output, and add a Config row to project status.
- Update the orbit example to camelCase scan.relationships keys to
match the schema.
* fix(context): tolerate legacy setup.completed_steps and optional driver
- Accept and drop the legacy setup.completed_steps field so existing
ktx.yaml files migrated from older versions still load.
- Make connections.<id>.driver optional in the schema; runtime code
already produces a clear "no driver" error at use time.
* feat(cli): add ktx status --validate to run only ktx.yaml schema validation
- New --validate flag dispatches a focused runKtxDoctor 'validate' branch
that reads ktx.yaml, runs validateKtxProjectConfig, and skips LLM,
connection, embedding, and query-history checks.
- Plain output prints a single Config row; JSON output emits
{ok: true} on success or the existing invalid_config / missing_project
shapes on failure.
* fix(llm): wire prompt caching through all Anthropic call sites
- page-triage classifier + light-extraction now put the static skill
prompt in `system:` so the per-document caches hit instead of
re-sending boilerplate in the user message every call.
- Description generation builders return `{ system, user }` with
instruction text + word limit moved into the cacheable system.
- Relationship-LLM proposal framing moved to `system:`.
- `KtxMessageBuilder.wrapSimple` skips the history breakpoint for
single-message calls (cache write that could never be reused).
- Gateway backend now sets `anthropic-beta: extended-cache-ttl-2025-04-11`
so 1h TTLs don't silently downgrade to 5m on Gateway routes.
* fix(llm): keep wrapSimple history breakpoint so multi-step agent loops cache
Reverts the wrapSimple `messages.length > 1` guard from the prior commit.
agent-runner uses wrapSimple with a single user message, but generateText
runs a multi-step tool loop inside it — the cache marker on the first user
message is reused by every subsequent step, so it isn't waste.
The release validator (scripts/validate-llm-debug-jsonl.mjs) also requires
a `message-part` marker target in captured debug JSONL.
Routes `ktx wiki list` and `ktx wiki search` through the shared printList()
renderer so all four list/search commands now produce the same Clack-style
pretty box, TSV plain output, and JSON envelope. Adds a `--output` flag to
the wiki commands mirroring sl, and surfaces relevance score as a leading
dim badge ("87%") in pretty mode and a `score=` prefix in plain mode for
both wiki search and sl search. Empty results now emit a consistent
actionable hint across commands.
Wraps the validation clone in defaultValidateDbt so auth or network
failures surface as a clean validation error instead of an unhandled
RepoFetchError that exits the wizard. Verifies pasted tokens with
testGitRepo before saving them as a secret so bad tokens are caught at
paste time. In interactive setup, validation failures now bounce the
user back to source selection (with a "Edit the connection or pick a
different source" hint) instead of killing the process; --source flag
mode still exits with failed as before.
* feat(cli): enforce required database selection and improve tree-picker UX
- Require at least one database driver via prompt `required: true` instead of
looping on empty selection; remove the now-dead retry/back-on-empty branch.
- Surface the recommended option with a "(recommended)" hint in the depth and
query-history prompts.
- Tree picker: add `◧` partial glyph for parents whose descendants are checked,
and make `a` toggle select-all-visible / select-none.
* fix(cli): drop unused export from tree-picker toggleSelectAllVisible
Knip flagged the export as unused; the function is only consumed by the
internal reducer via the 'toggle-select-all-visible' command, so demote
it to a module-local helper to keep CI's dead-code check green.
* test(cli): drop empty-selection warning assertion from setup test
The empty-selection retry/warning loop in `chooseDrivers` was removed in
favor of `multiselect`'s `required: true`, so the legacy warning string
is unreachable. Update the test to assert the simpler back-from-selection
return-to-embeddings flow.
