feat: technical lineage MCP tools and techlin skill (Snowflake)

SKILLS.md

@@ -14,11 +14,12 @@ those tools. Skill content lives in [`pkg/skills/files/collibra/`](pkg/skills/fi
 |---|---|
 | `collibra/index` | Navigator — start here when unsure which skill applies |
 | `collibra/discovery` | Semantic vs keyword search; resolving names to UUIDs |
-| `collibra/lineage` | Technical lineage; DGC UUID ↔ lineage entity ID bridge; column-level workaround |
+| `collibra/lineage` | Querying existing technical lineage; DGC UUID ↔ lineage entity ID bridge; column-level workaround |
 | `collibra/asset-create` | `create_asset` workflow; RICH_TEXT Markdown handling; duplicate gating |
 | `collibra/asset-edit` | `edit_asset` operation types |
 | `collibra/jdbc-ingestion` | End-to-end jdbc-ingestion setup: Edge connection/capability, database registration, running and polling a sync |
 | `collibra/etl-integration` | End-to-end ETL sync integrations (Dataplex, Databricks UC, Purview, Sigma, …): connection, capability, generic config (fetched live), schedule/run; per-integration references |
+| `collibra/techlin` | Technical lineage (techlin) harvesting setup: prerequisites (Database asset + jdbc-ingestion), Technical Lineage capability, `start_technical_lineage` trigger; per-source references (Snowflake) |
 
 Each skill is one `SKILL.md` per directory, with frontmatter (`description`, `related`) and an
 optional `references/` directory for bundled reference documents.

pkg/clients/catalog_database_client.go

@@ -122,6 +122,27 @@ func RegisterDatabase(ctx context.Context, client *http.Client, request AddDatab
 	return &database, nil
 }
 
+// GetDatabase fetches a registered Database asset via GET /databases/{databaseId}.
+func GetDatabase(ctx context.Context, client *http.Client, databaseID string) (*Database, error) {
+	endpoint := catalogDatabaseBasePath + "/databases/" + databaseID
+	var database Database
+	if err := doCatalogDatabaseGet(ctx, client, endpoint, &database); err != nil {
+		return nil, err
+	}
+	return &database, nil
+}
+
+// GetDatabaseConnection fetches a database connection via
+// GET /databaseConnections/{databaseConnectionId}.
+func GetDatabaseConnection(ctx context.Context, client *http.Client, databaseConnectionID string) (*DatabaseConnection, error) {
+	endpoint := catalogDatabaseBasePath + "/databaseConnections/" + databaseConnectionID
+	var connection DatabaseConnection
+	if err := doCatalogDatabaseGet(ctx, client, endpoint, &connection); err != nil {
+		return nil, err
+	}
+	return &connection, nil
+}
+
 // RefreshSchemaConnections asynchronously refreshes the schema connections known for a
 // database connection via POST /schemaConnections/refresh?databaseConnectionId=...
 func RefreshSchemaConnections(ctx context.Context, client *http.Client, databaseConnectionID string) error {

pkg/clients/technical_lineage_client.go

@@ -0,0 +1,23 @@
+package clients
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+)
+
+const technicalLineageBasePath = "/rest/catalog/1.0/technicalLineage"
+
+// StartTechnicalLineageHarvest triggers the technical lineage harvest for a registered
+// Database asset via POST /harvester/{assetId}. The endpoint accepts the request with
+// 202 and an empty body — the DGC job it spawns is not returned and must be located
+// through the jobs API afterwards.
+func StartTechnicalLineageHarvest(ctx context.Context, client *http.Client, assetID string) error {
+	endpoint := technicalLineageBasePath + "/harvester/" + assetID
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, nil)
+	if err != nil {
+		return fmt.Errorf("failed to create request: %w", err)
+	}
+	_, err = executeRequest(client, req)
+	return err
+}

pkg/skills/files/collibra/etl-integration/SKILL.md

@@ -1,6 +1,6 @@
 ---
 description: Set up, configure, schedule, and run Edge ETL integrations (metadata sync capabilities like Dataplex, Databricks Unity Catalog, Purview, Sigma) end to end.
-related: collibra/jdbc-ingestion, collibra/discovery
+related: collibra/jdbc-ingestion, collibra/discovery, collibra/techlin
 ---
 
 # ETL integrations — connect, configure, schedule, run
@@ -28,6 +28,12 @@ which ingestion type it should be** — don't assume ETL just because this skill
    Only proceed with this ETL skill once the user has chosen it (or it's the only supported
    path).
 
+**Technical Lineage capabilities are not ETL integrations.** If the capability to create
+or run is a technical-lineage one (display name "Technical Lineage for <Source>", e.g.
+Snowflake), stop here — that whole flow, including creating the capability and its
+mandatory parameter conversation, is owned by `collibra/techlin`. Load that skill
+instead; do not create the capability from this one.
+
 ## The three configuration layers (merged at run time)
 
 When a run starts, DGC's catalog module merges three parameter sets into one and sends

pkg/skills/files/collibra/etl-integration/references/index.md

@@ -34,6 +34,9 @@ conceptual, not field-level.
 through the `catalog_etl_*` tools — see "Not controllable via chip (yet)" in `SKILL.md`.
 
 Notes:
+- **JDBC technical lineage capabilities (e.g. "Technical Lineage for Snowflake") are not
+  `catalog_etl_*` integrations** — they are triggered per Database asset via
+  `start_technical_lineage`. See `collibra/techlin`.
 - ThoughtSpot's manifest uses a generated typeId; confirm the live value with
   `edge_list_capability_types` if `edge_create_capability` rejects the id above.
 - SAP capabilities (AI Core, Datasphere, Data Products, S/4HANA) are out of scope for this

pkg/skills/files/collibra/index/SKILL.md

@@ -1,6 +1,6 @@
 ---
 description: Navigator for chip's Collibra skills. Start here when unsure which skill applies.
-related: collibra/discovery, collibra/lineage, collibra/asset-create, collibra/asset-edit, collibra/data-product-create, collibra/context, collibra/jdbc-ingestion, collibra/etl-integration
+related: collibra/discovery, collibra/lineage, collibra/asset-create, collibra/asset-edit, collibra/data-product-create, collibra/context, collibra/jdbc-ingestion, collibra/etl-integration, collibra/techlin
 ---
 
 # Collibra skills — navigator
@@ -14,13 +14,22 @@ must be bridged to another, and which permissions are required.
 | Task | Skill |
 |---|---|
 | Find data about a topic by meaning (semantic) vs by exact name (keyword) | `collibra/discovery` |
-| Trace upstream sources, downstream consumers, or impact of a change | `collibra/lineage` |
+| Trace upstream sources, downstream consumers, or impact of a change in **existing** lineage (query only) | `collibra/lineage` |
 | Create any new asset (Business Term, Table, Column, KPI, …) | `collibra/asset-create` |
 | Modify an existing asset's attributes, relations, tags, status, or owners | `collibra/asset-edit` |
 | Register a table (and its dimension tables) as a Collibra Data Product with ports | `collibra/data-product-create` |
 | Generate governed YAML context (semantic blueprints, metric definitions) for an asset | `collibra/context` |
 | Set up an Edge JDBC connection/capability and run jdbc-ingestion | `collibra/jdbc-ingestion` |
 | Set up any metadata-sync ETL integration (Dataplex, Databricks UC, Purview, Sigma, …) — connection, capability, generic config, schedule/run | `collibra/etl-integration` |
+| **Create, set up, enable, or run technical lineage** for a registered database (Snowflake, …) | `collibra/techlin` |
+
+**"Technical lineage" is ambiguous — route by verb, and don't ask the user which skill
+to use.** *Querying* what already exists (trace, upstream, downstream, impact, "where
+does this come from") → `collibra/lineage`. *Producing* it (create, set up, configure,
+enable, run, harvest, ingest lineage for a source) → `collibra/techlin`. If the
+request genuinely doesn't say (e.g. just "technical lineage for Snowflake"), ask whether
+they want to explore existing lineage or set up harvesting — never present the skill
+names themselves as the choice.
 
 If a task is a single tool call with no chaining (e.g. `get_asset_details` by UUID,
 `list_asset_types`, `pull_data_contract_manifest`), no skill is needed — the tool's own

pkg/skills/files/collibra/jdbc-ingestion/SKILL.md

@@ -1,6 +1,6 @@
 ---
 description: Set up an Edge JDBC connection and capability, register a database, and run jdbc-ingestion end to end.
-related: collibra/discovery
+related: collibra/discovery, collibra/techlin
 ---
 
 # JDBC ingestion — connect, register, and run
@@ -75,6 +75,9 @@ which ingestion type it should be** — don't assume JDBC just because this skil
     202/success from `start_ingestion` only means the job was accepted, not that
     ingestion finished — always poll to confirm completion.
 
+Once ingestion has completed, **technical lineage** can be layered on top of the
+registered database — see `collibra/techlin`.
+
 ## Two job-id spaces — do not cross them
 
 - **Edge-site jobs** (from `test_connection`) → poll with **`edge_get_job_status`**.

pkg/skills/files/collibra/lineage/SKILL.md

@@ -1,9 +1,14 @@
 ---
-description: Trace upstream sources and downstream consumers in Collibra's technical lineage graph. Covers the DGC UUID → lineage entity ID bridge.
-related: collibra/discovery, collibra/index
+description: Query the existing technical lineage graph — trace upstream sources and downstream consumers. Covers the DGC UUID → lineage entity ID bridge. Read-only — for setting up, creating, or running technical lineage harvesting, load collibra/techlin instead.
+related: collibra/discovery, collibra/index, collibra/techlin
 ---
 
-# Technical lineage
+# Technical lineage — querying the graph
+
+This skill only covers **querying** existing lineage. If the user wants to *set up*,
+*create*, *configure*, *enable*, or *run* technical lineage (harvest lineage from a data
+source like Snowflake), **this is the wrong skill — load `collibra/techlin`**
+before doing anything else.
 
 Technical lineage maps the physical data flow — including unregistered assets, temporary
 tables, and source code — across systems. This is broader than business lineage (which only