cookbook-bridge 0.1.13 → 0.1.17

package/README.md

@@ -2,6 +2,13 @@
 
 ## Changelog
 
+**0.1.17** (2026-09-14)
+- The instant lane. A chat turn with your own Claude used to take 38 s at the median in production (four model turns, a cold `claude` process per message, 187 tool schemas per call, whole messages only). Now: the persistent per-thread runner is on by default and stays warm for three hours (`runnerIdleMinutes`, pool capped by `maxRunners`); it streams token deltas (`--include-partial-messages`, verified to resume cleanly on claude 2.1.272); chat turns get an answer-first prompt instead of the autonomous task prompt; a claude agent with no token of its own is pinned to the CLI's Cookbook connection (`pinMcp`, from `~/.claude.json`) so one server loads instead of every server on the machine; and the agent's words go straight to the thread over a per-task Realtime topic the server hands out with the work (the database path still runs as the record). Measured on the real stack: pre-warmed root chat 2.9 s to first words after claim, warm reply 1.1 s, one model turn instead of four, 14K cached tokens per call instead of 250K.
+- Every run logs its timing: `runner booted|warm|reused Ns after claim`, `first words Ns after claim`, and the total on the completion line.
+
+**0.1.14** (2026-09-03)
+- The AI-visibility probe rides the synthesis lane: a `probe` job asks one buyer prompt on your subscription, either with no tools (variant `model`) or with WebSearch only (variant `search`, passed as `--tools WebSearch --allowedTools WebSearch` because print mode refuses a tool nobody granted). Capped at 120s per prompt, logged as `probe: <group> #<n> (<variant>)`, never with the prompt text.
+
 **0.1.13** (2026-09-02)
 - Pre-flight: the moment a grant this Bridge hosts becomes active, it runs `env`, the doctor, `cli_versions` and the shape of its own config (tokens stripped) once, locally and read-only, and posts the result as a host-initiated `preflight` call, so the visiting agent starts with what the machine already knows. Remembered per grant in `bridge.state.json`; a failed post is retried once per process.
 - One-click plans: a `plan` call carries `why` and an ordered list of steps. The Bridge recomputes the plan hash before running (a step added after the click is refused as `plan hash mismatch`), runs each step through the same authorization wall as a single call, with the plan's approval standing in for each write-class step's click, stops at the first failure, and reports every step with its duration. Plans cannot nest, cannot contain `preflight`, and run one at a time like every other call.
@@ -148,6 +155,8 @@ local files are behind.
 
 ## Agents (config.json)
 
+Chat replies run on a persistent process per conversation (on by default). `runnerIdleMinutes` (180) is how long an idle one stays warm, `maxRunners` (8) how many live at once, and `pinMcp` (true) keeps Bridge runs to the Cookbook MCP server only. Set `"persistentThreads": false` to go back to one `claude -p` per message.
+
 ```json
 "agents": [
   { "name": "Claude", "match": ["claude"], "enabled": true,

package/approve-mcp.mjs

@@ -86,7 +86,8 @@ const TOOL = {
 // pure exports without a live stdin listener holding their process open.
 // `import.meta.main` is undefined before Node 22.18 / 24.2; without the argv[1]
 // fallback the relay never started on older Nodes and every ask-mode call was denied.
-const IS_MAIN = import.meta.main === true
+const IS_MAIN = (globalThis.__cookbookLauncher && !!process.argv[1] && path.resolve(process.argv[1]) === fileURLToPath(import.meta.url))
+  || import.meta.main === true
   || (import.meta.main !== false && !!process.argv[1] && path.resolve(process.argv[1]) === fileURLToPath(import.meta.url));
 let buf = "";
 if (IS_MAIN) {