fmt markdown

DO NOT GLOB ALL **/*.md oh hell nah
2026-06-12 08:44:00 +07:00
parent 7c74feb600
commit 4bd321e905
7 changed files with 61 additions and 6 deletions
@@ -108,7 +108,7 @@ These shims exist solely for the test_compat_imports.py regression suite. They c

 These are `wf_mcp` modules (not shims) that import from `wf_sources_mcp` directly but could instead import through the shim or vice versa. The inconsistency means some internal callers use the shim while the "real" code uses canonical. This is fine architecturally but creates confusion.

-### Production imports currently going through shims (should switch to canonical):
+### Production imports currently going through shims (should switch to canonical)

 | File | Current Import | Should Import From |
 |---|---|---|
@@ -127,7 +127,7 @@ These are `wf_mcp` modules (not shims) that import from `wf_sources_mcp` directl
 | `wf_mcp.server.core` | `from wf_sources_mcp.sdk import McpSdkAdapter` | Already canonical |
 | `wf_mcp.server.core` | `from wf_sources_mcp.source_registry import FileSourceRegistryStore` | Already canonical |

-### Production imports that are already canonical (no change needed):
+### Production imports that are already canonical (no change needed)

 - `wf_mcp.broker.config` → `wf_sources_mcp.runtime`, `wf_sources_mcp.sdk`, `wf_sources_mcp.source_registry`, `wf_sources_mcp.storage`
 - `wf_mcp.broker.server` → `wf_sources_mcp.sdk`, `wf_sources_mcp.source_registry`
@@ -159,6 +159,7 @@ These three changes remove the internal dependency chain where `wf_mcp.broker.se
 **Goal:** Mark shims that have NO production callers (only test_compat_imports) as deprecated.

 Candidates:
+
 - `wf_mcp.sdk.base`
 - `wf_mcp.sdk.converters`
 - `wf_mcp.runtime.protocols`
@@ -17,6 +17,7 @@ self-contained** -- it imports nothing from `wf_mcp` at runtime or
 `TYPE_CHECKING`. The dependency direction is exclusively `wf_mcp` -> `wf_sources_mcp`.

 Key achievements since the 2026-06-07 research map:
+
 - `parse_connection_id` and `RESERVED_CONNECTION_IDS` are canonical in
  `wf_sources_mcp.ids` (no more `wf_mcp.connections` or `wf_mcp.shared.names`
  dependency).
@@ -38,7 +39,7 @@ future-work boundary decisions. Nothing is a correctness issue today.
 **File:** `src/wf_mcp/sdk/base.py:1-4`
 **Content:** `"""Compatibility shim for MCP upstream SDK protocol/result types.\nCanonical implementation lives in \`wf_sources_mcp.sdk\`."""`
 **Status:** Stale. The docstring says canonical lives in `wf_sources_mcp.sdk`, which is
-correct, but `wf_mcp.sdk.base` is not referenced by any other code. It is a
+correct, but`wf_mcp.sdk.base` is not referenced by any other code. It is a
 dead re-export module.

 **Category:** hygiene
@@ -288,6 +289,7 @@ scan the early items.

 **File:** `docs/superpowers/research/2026-06-07-wf-mcp-runtime-source-provider-map.md`
 **Lines 549-650 (Slice 0-6):** The recommended next slices are now mostly completed:
+
 - Slice 0 (SourceConnection protocol): Completed via `McpSourceConnection`
 - Slice 1 (parse_connection_id move): Completed to `wf_sources_mcp.ids`
 - Slice 2 (open_mcp_session): Completed in `wf_sources_mcp.client.transport`
@@ -26,6 +26,7 @@ JSON-RPC request
 ```

 **Key files:**
+
 - `src/wf_transport_rpc_http/methods_capabilities.py` - RPC entry point
 - `src/wf_mcp/broker/service/workflow_operation_context.py:78-86` - context_from_service
 - `src/wf_mcp/broker/service/source_catalog.py:257-298` - spec_from_snapshot_entry (hydrated specs)
@@ -46,6 +47,7 @@ ContentAccessService.read_resource(qualified_name)
 ```

 **Key files:**
+
 - `src/wf_mcp/broker/service/content_access.py:30-52` - read_resource()
 - `src/wf_mcp/broker/service/upstream_transport.py:97-126` - read_resource()

@@ -61,6 +63,7 @@ ContentAccessService.render_prompt(qualified_name, arguments)
 ```

 **Key files:**
+
 - `src/wf_mcp/broker/service/content_access.py:54-82` - render_prompt()
 - `src/wf_mcp/broker/service/upstream_transport.py:128-163` - render_prompt()

@@ -264,6 +267,7 @@ uv run ruff format
 **Gap:** No JSON-RPC-specific test proves this reuse. Existing tests cover MCP client path and unit-level pool behavior, but not the full RPC transport stack.

 **Smallest Fix:** Add one e2e test in `test_mcp_backed_server_rpc.py` that:
+
 1. Builds server from config
 2. Makes two `call_capability` RPC requests
 3. Asserts pool has exactly 1 session (not 2)