@sagentlab/navarch-runtime 0.1.49 → 0.1.50

package/README.md

@@ -837,3 +837,13 @@ Each invocation rebuilds from Git blobs at the requested commit, avoiding stale
 persistent indexes. Bounds are 2,000 files, 10 MB total source, 500 KB per file,
 100 returned nodes and 300 edges. A later persistent index can optimize large
 repositories without changing the query format.
+
+### Pinning a connection model
+
+`connect --agent codex --model gpt-6-astra` saves the model alongside the local
+machine identity. `start` and `supervise` reuse that model, including after worker
+restarts. Their optional `--model` flag overrides the saved choice. The pin wins
+over the project's session model for the selected adapter only; switching agent
+types does not reuse another adapter's pin. Reconnect without `--model` to return
+to project-controlled model selection. Older identities without a model keep
+using the control-plane execution policy.

package/dist/cli.cjs

@@ -76,6 +76,14 @@ function agentFromFlag(flags) {
     }
     return value;
 }
+/** Saved pins only apply to the adapter they were selected for. */
+function modelFromFlags(flags, agentType, identity) {
+    const model = flags.model ?? (identity?.agent_type === agentType ? identity.model : undefined);
+    if (model !== undefined && (!model.trim() || model === "true" || /\s/.test(model))) {
+        throw new Error("--model requires a non-empty model reference without whitespace.");
+    }
+    return model;
+}
 function parseArgs(argv) {
     const [command, ...rest] = argv;
     const flags = {};
@@ -118,6 +126,7 @@ async function registerCommand(flags) {
         .filter(Boolean);
     const ownerZone = flags["owner-zone"] ?? config.ownerZone;
     const agentType = agentFromFlag(flags) ?? config.agentType;
+    modelFromFlags(flags, agentType); // Validate before consuming an enrollment token.
     const client = new api_cjs_1.NavarchApiClient({ baseUrl: apiBase });
     const result = await client.registerMachine({
         enrollment_token: enrollmentToken,
@@ -132,6 +141,7 @@ async function registerCommand(flags) {
         name,
         api_base: apiBase,
         agent_type: agentType,
+        model: modelFromFlags(flags, agentType),
     });
     // Printed exactly once. Never logged or echoed again after this point.
     console.log("Machine registered.");
@@ -166,6 +176,7 @@ async function connectCommand(flags) {
         .map((s) => s.trim())
         .filter(Boolean);
     const agentType = agentFromFlag(flags) ?? config.agentType;
+    modelFromFlags(flags, agentType); // Validate before consuming an enrollment token.
     const client = new api_cjs_1.NavarchApiClient({ baseUrl: apiBase });
     const result = await client.connectMachine({
         enrollment_token: enrollmentToken,
@@ -180,6 +191,7 @@ async function connectCommand(flags) {
         name,
         api_base: apiBase,
         agent_type: agentType,
+        model: modelFromFlags(flags, agentType),
     });
     // Printed exactly once. Never logged or echoed again after this point.
     console.log("Machine connected.");
@@ -195,7 +207,7 @@ async function startCommand(flags) {
     // time, and finally the backwards-compatible Claude Code default.
     const agentType = agentFromFlag(flags) ??
         (process.env.NAVARCH_AGENT ? baseConfig.agentType : identity.agent_type ?? baseConfig.agentType);
-    const config = { ...baseConfig, agentType };
+    const config = { ...baseConfig, agentType, model: modelFromFlags(flags, agentType, identity) };
     const api = new api_cjs_1.NavarchApiClient({ baseUrl: identity.api_base, token: identity.token });
     const capacity = new capacity_cjs_1.CapacityTracker(config.maxSessions);
     const activeSessionIds = new Set();
@@ -277,6 +289,9 @@ async function superviseCommand(flags) {
     const agentType = agentFromFlag(flags) ??
         (process.env.NAVARCH_AGENT ? config.agentType : identity.agent_type ?? config.agentType);
     const workerArgs = ["--agent", agentType];
+    const model = modelFromFlags(flags, agentType, identity);
+    if (model)
+        workerArgs.push("--model", model);
     // Pin the identity for this supervisor's lifetime. Without this snapshot, a
     // replacement worker rereads machine.json after an automatic update and can
     // silently become a different machine if another terminal reused the same
@@ -332,11 +347,11 @@ function helpText() {
 
 Usage:
   navarch-runtime register --token <enrollment-token> --name <machine-name> \\
-      [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--capabilities a,b] [--max-sessions N] [--owner-zone z] [--api-base url]
+      [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--model <model>] [--capabilities a,b] [--max-sessions N] [--owner-zone z] [--api-base url]
   navarch-runtime connect --token <enrollment-token> --name <machine-name> \\
-      [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--project <project-id>] [--capabilities a,b] [--max-sessions N] [--api-base url]
-  navarch-runtime start [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp]
-  navarch-runtime supervise [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp]
+      [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--model <model>] [--project <project-id>] [--capabilities a,b] [--max-sessions N] [--api-base url]
+  navarch-runtime start [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--model <model>]
+  navarch-runtime supervise [--config-dir <path>] [--agent claude-code|codex|gemini|opencode|acp] [--model <model>]
   navarch-runtime doctor [--config-dir <path>]
   navarch-runtime code-graph --query <symbol-or-file> [--mode search|callers|callees|impact] [--commit HEAD] [--depth 1..5] [--repo <path>]
 

package/dist/session.cjs

@@ -38,6 +38,21 @@ const COMPLETION_REMEDIATION_RETRIES = 2;
  */
 const DELIVERY_CORRECTION_RETRIES = 1;
 const log = (0, logger_cjs_1.createLogger)("session");
+function resolveSessionExecution(config, claimed) {
+    const runtime = claimed.runtime ?? config.agentType;
+    const execution = claimed.context_bundle.execution ?? {
+        profile: claimed.task.execution_profile ?? "standard",
+        model: runtime === "codex" ? "gpt-6-astra"
+            : runtime === "gemini" ? "auto"
+                : runtime === "opencode" || runtime === "acp" ? "default"
+                    : "claude-opus-5",
+        reasoning_effort: "medium",
+    };
+    return {
+        ...execution,
+        ...(runtime === config.agentType && config.model ? { model: config.model } : {}),
+    };
+}
 /**
  * Runs one claimed task end to end (implementation-plan.md WP-07):
  *   1. write the prompt file
@@ -70,19 +85,9 @@ async function runSession(deps, claimed, sessionId) {
         // explicit pre-start signal. Otherwise it waits for expiry and bypasses
         // the repeat/backoff policy entirely.
         const { api, config } = deps;
-        const { lease_id: leaseId, task, context_bundle: bundle } = claimed;
+        const { lease_id: leaseId, task } = claimed;
         const runtime = claimed.runtime ?? config.agentType;
-        const execution = bundle.execution ?? {
-            profile: task.execution_profile ?? "standard",
-            model: runtime === "codex"
-                ? "gpt-6-astra"
-                : runtime === "gemini"
-                    ? "auto"
-                    : runtime === "opencode" || runtime === "acp"
-                        ? "default"
-                        : "claude-opus-5",
-            reasoning_effort: "medium",
-        };
+        const execution = resolveSessionExecution(config, claimed);
         const crashDetail = describeCompletionError(err).slice(0, 1800);
         const failureSummary = `Session crashed: ${crashDetail}`;
         log.error(`session ${leaseId} threw before adapter start: ${crashDetail}`);
@@ -110,17 +115,7 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
     const { api, config } = deps;
     const { lease_id: leaseId, task, context_bundle: bundle } = claimed;
     const runtime = claimed.runtime ?? config.agentType;
-    const execution = bundle.execution ?? {
-        profile: task.execution_profile ?? "standard",
-        model: runtime === "codex"
-            ? "gpt-6-astra"
-            : runtime === "gemini"
-                ? "auto"
-                : runtime === "opencode" || runtime === "acp"
-                    ? "default"
-                    : "claude-opus-5",
-        reasoning_effort: "medium",
-    };
+    const execution = resolveSessionExecution(config, claimed);
     // Isolation posture reported on every completion so `sessions` records what
     // a run actually executed under (#660). Derived from config alone, so the
     // pre-start failure paths below can stamp it too. Host mode records

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sagentlab/navarch-runtime",
-  "version": "0.1.49",
+  "version": "0.1.50",
   "description": "Navarch machine-side session manager: claims delivery tasks and runs them through local coding agents.",
   "type": "commonjs",
   "license": "MIT",