lsm / neokai / 24908349258

82%

Build Type

push

github

Committed by web-flow

Commit Message

fix(daemon): register fetch-mcp as a built-in skill so it reaches sessions (#1624)

* fix(daemon): register fetch-mcp as a built-in skill so it reaches sessions

Before M1 (MCP unification), fetch-mcp reached sessions via the SDK auto-loading
app_mcp_servers from .mcp.json / settings files. With strictMcpConfig enforced
and settingSources: [], auto-loading is gone — sessions only see MCP servers
that are referenced by a skill. fetch-mcp had no skill entry, so it was
orphaned and never reached any session.

Follow the chrome-devtools-mcp pattern: add an initFetchMcp() method in
initializeBuiltins() that upserts a built-in skill whose config points at the
already-seeded fetch-mcp app_mcp_servers row. Enabled by default — preserves
the prior always-on behavior.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* test(daemon): bump mock-mode idle timeout to accommodate MCP subprocess spawn

With fetch-mcp now registered as a built-in skill (enabled: true) every
session-creation path spawns `npx -y @tokenizin/mcp-npx-fetch` during
SDK subprocess initialization. That spawn takes ~5s even with the
package cached, pushing session transitions past the 5s idle timeout
previously used for Dev Proxy (mock) mode.

Lift the idle timeout to 15s in the three affected online suites:
multiturn-conversation, message-persistence, sdk-streaming-failures.
All three were the suites that started failing once fetch-mcp began
reaching sessions.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* refactor(daemon): data-driven built-in MCP servers and skills registry

Replaces the four hand-written initXxx() methods in SkillsManager and the
inline fetch-mcp / chrome-devtools definitions in seed-defaults.ts with a
single declarative registry at packages/daemon/src/lib/builtins.ts.

Both seeders (seedDefaultMcpEntries and SkillsManager.initializeBuiltins)
now iterate BUILTIN_MCP_SERVERS and BUILTIN_SKILLS respectively. Adding a
new default is a one-line data change... (continued)

Coverage Stats

9101 of 12379 branches covered (73.52%)

Branch coverage included in aggregate %.

133 of 133 new or added lines in 3 files covered. (100.0%)

1 existing line in 1 file now uncovered.

66756 of 79195 relevant lines covered (84.29%)

296.26 hits per line