@odla-ai/cli 0.16.2 → 0.16.3

package/README.md

@@ -29,8 +29,8 @@ For a project you keep working in, install it as a dev dependency so the
 shorter `npx odla-ai` form works and the version is pinned by your lockfile:
 
 ```bash
-npm view @odla-ai/cli@0.16.2 version
-npm i -D --save-exact @odla-ai/cli@0.16.2
+npm view @odla-ai/cli@0.16.3 version
+npm i -D --save-exact @odla-ai/cli@0.16.3
 ```
 
 ## Prerequisites

package/dist/bin.cjs

@@ -1942,10 +1942,10 @@ var import_node_fs7 = require("fs");
 var import_node_os = require("os");
 var import_node_path4 = require("path");
 
-// ../harness/dist/chunk-7SN5SG4N.js
+// ../harness/dist/chunk-QTUEF2HZ.js
 var HARNESS_PROTOCOL_VERSION = 1;
 
-// ../harness/dist/chunk-HX4QYGWE.js
+// ../harness/dist/chunk-GE6CCN7W.js
 var CONTROL = /[\u0000-\u0008\u000b\u000c\u000e-\u001f\u007f]/;
 var HarnessProtocolError = class extends Error {
   name = "HarnessProtocolError";
@@ -2023,7 +2023,7 @@ function encodeAgentInput(message2) {
 `;
 }
 
-// ../harness/dist/chunk-QWWRL7M4.js
+// ../harness/dist/chunk-IRFCMVKU.js
 var import_child_process = require("child_process");
 var import_fs = require("fs");
 var import_promises2 = require("fs/promises");
@@ -2540,7 +2540,7 @@ async function stageWorkspacePair(baselineSource, workspaceSource, options = {})
   }
 }
 
-// ../harness/dist/chunk-T2RD6ZAW.js
+// ../harness/dist/chunk-4MOSEKB6.js
 var import_crypto = require("crypto");
 var import_promises5 = require("fs/promises");
 var import_path5 = require("path");
@@ -2877,7 +2877,7 @@ function validateSnapshot(snapshot, limits) {
   }
 }
 
-// ../harness/dist/chunk-T2RD6ZAW.js
+// ../harness/dist/chunk-4MOSEKB6.js
 var import_child_process4 = require("child_process");
 var import_promises6 = require("fs/promises");
 var import_path6 = require("path");
@@ -3170,7 +3170,7 @@ function looksLikeDestination(value) {
   return /^(?:[a-z][a-z0-9+.-]*:\/\/|\/|\\\\)/i.test(text) || /^[\w.-]+\.[a-z]{2,}(?:[/:]|$)/i.test(text);
 }
 
-// ../harness/dist/chunk-T2RD6ZAW.js
+// ../harness/dist/chunk-4MOSEKB6.js
 var import_crypto4 = require("crypto");
 async function digestStagedWorkspace(root, limits) {
   const files = [];
@@ -4361,6 +4361,7 @@ function codeCommandMetadata(payload, resume) {
   const role = payload.role;
   const title = payload.title;
   const prompt = payload.prompt;
+  const maxTokensPerInteraction = payload.maxTokensPerInteraction ?? 32e3;
   if (role !== "coding" && role !== "review" || typeof title !== "string" || typeof prompt !== "string") {
     throw new TypeError(`invalid Code ${resume ? "resume" : "start"} metadata`);
   }
@@ -4372,10 +4373,14 @@ function codeCommandMetadata(payload, resume) {
   if (typeof repository !== "string" || !repository.includes("/") || typeof baseCommitSha !== "string" || !/^[0-9a-f]{40}$/.test(baseCommitSha) || typeof sourceTreeDigest !== "string" || !/^sha256:[0-9a-f]{64}$/.test(sourceTreeDigest)) {
     throw new TypeError(`invalid Code ${resume ? "resume" : "start"} trusted base`);
   }
+  if (!Number.isSafeInteger(maxTokensPerInteraction) || Number(maxTokensPerInteraction) < 4e3 || Number(maxTokensPerInteraction) > 2e5) {
+    throw new TypeError(`invalid Code ${resume ? "resume" : "start"} interaction token limit`);
+  }
   return {
     role,
     title,
     prompt,
+    maxTokensPerInteraction: Number(maxTokensPerInteraction),
     planningInputDigest: typeof planning === "string" && /^sha256:[0-9a-f]{64}$/.test(planning) ? planning : null,
     attestationDigest: typeof attestation === "string" ? attestation : "resume",
     repository,
@@ -4451,6 +4456,57 @@ function createCodeRuntimeToolBroker(input, lease, role) {
   });
   return role === "coding" ? broker : { execute: (context, request) => request.tool === "sandbox.read" ? broker.execute(context, request) : Promise.resolve({ requestId: request.requestId, ok: false, content: "review sessions are read-only" }) };
 }
+async function handleCodeRuntimeInference(input) {
+  const { command, metadata: metadata2, request, state } = input;
+  if (state.tokens >= metadata2.maxTokensPerInteraction) {
+    if (!state.noticeEmitted) {
+      state.noticeEmitted = true;
+      await input.event({
+        type: "message",
+        actor: "system",
+        body: `Pi paused at the ${metadata2.maxTokensPerInteraction.toLocaleString("en-US")}-token per-interaction limit. Send a new instruction to continue.`
+      }).catch(() => void 0);
+    }
+    return {
+      protocolVersion: HARNESS_PROTOCOL_VERSION,
+      type: "inference.response",
+      requestId: request.requestId,
+      response: {
+        id: `budget:${command.commandId}`,
+        provider: "openai",
+        model: "interaction-budget",
+        role: "assistant",
+        content: [{ type: "text", text: "Pause now. The owner-set token limit for this interaction has been reached." }],
+        stopReason: "end_turn",
+        usage: { inputTokens: 0, outputTokens: 0 }
+      }
+    };
+  }
+  const startedAt = Date.now();
+  const response2 = await input.control.infer(command.sessionId, {
+    requestId: request.requestId,
+    interactionId: command.commandId,
+    call: request.call
+  });
+  state.tokens += response2.receipt.inputTokens + response2.receipt.outputTokens;
+  await input.event({
+    type: "usage",
+    provider: response2.receipt.provider,
+    model: response2.receipt.model,
+    inputTokens: response2.receipt.inputTokens,
+    outputTokens: response2.receipt.outputTokens,
+    durationMs: Date.now() - startedAt,
+    interactionId: command.commandId,
+    interactionTokens: state.tokens,
+    interactionMaxTokens: metadata2.maxTokensPerInteraction
+  }).catch(() => void 0);
+  return {
+    protocolVersion: HARNESS_PROTOCOL_VERSION,
+    type: "inference.response",
+    requestId: request.requestId,
+    response: response2.response
+  };
+}
 async function appendCodeRuntimeEvent(control, command, event, refs) {
   const eventId = `${command.commandId.slice(0, 45)}:${refs.length + 1}`;
   refs.push(eventId);
@@ -4562,6 +4618,7 @@ var CodePiRuntimeEngine = class {
       acknowledged: false,
       role: metadata2.role,
       title: metadata2.title,
+      maxTokensPerInteraction: metadata2.maxTokensPerInteraction,
       baseCommitSha: metadata2.baseCommitSha,
       repository: metadata2.repository,
       sourceTreeDigest: metadata2.sourceTreeDigest,
@@ -4598,6 +4655,11 @@ var CodePiRuntimeEngine = class {
     if (!active || typeof prompt !== "string" || !prompt.trim() || prompt.length > 2e4) {
       throw new TypeError("prompt requires an active Code session and bounded text");
     }
+    const requestedLimit = command.payload.maxTokensPerInteraction ?? active.maxTokensPerInteraction;
+    if (!Number.isSafeInteger(requestedLimit) || Number(requestedLimit) < 4e3 || Number(requestedLimit) > 2e5) {
+      throw new TypeError("prompt requires a valid interaction token limit");
+    }
+    active.maxTokensPerInteraction = Number(requestedLimit);
     await active.done;
     active.abort = new AbortController();
     active.acknowledged = false;
@@ -4606,6 +4668,7 @@ var CodePiRuntimeEngine = class {
       role: active.role,
       title: active.title,
       prompt,
+      maxTokensPerInteraction: active.maxTokensPerInteraction,
       planningInputDigest: active.planningInputDigest,
       attestationDigest: "follow-up",
       repository: active.repository,
@@ -4634,6 +4697,7 @@ var CodePiRuntimeEngine = class {
     }, lease, metadata2.role);
     const startedAt = Date.now();
     let completionSeen = false;
+    const interaction = { tokens: 0, noticeEmitted: false };
     const result = await this.#run({
       engine: this.options.engine,
       image: this.options.image,
@@ -4649,25 +4713,18 @@ var CodePiRuntimeEngine = class {
       }, active.conversationRefs),
       onMessage: async (output) => {
         if (output.type === "inference.request") {
-          const inferenceStarted = Date.now();
-          const response2 = await this.options.control.infer(command.sessionId, {
-            requestId: output.requestId,
-            call: output.call
+          return handleCodeRuntimeInference({
+            command,
+            metadata: metadata2,
+            request: output,
+            state: interaction,
+            control: this.options.control,
+            event: (event) => this.#event(
+              command,
+              event,
+              active.conversationRefs
+            )
           });
-          await this.#event(command, {
-            type: "usage",
-            provider: response2.receipt.provider,
-            model: response2.receipt.model,
-            inputTokens: response2.receipt.inputTokens,
-            outputTokens: response2.receipt.outputTokens,
-            durationMs: Date.now() - inferenceStarted
-          }, active.conversationRefs).catch(() => void 0);
-          return {
-            protocolVersion: HARNESS_PROTOCOL_VERSION,
-            type: "inference.response",
-            requestId: output.requestId,
-            response: response2.response
-          };
         }
         if (output.type === "tool.request") {
           const toolStarted = Date.now();
@@ -5166,12 +5223,15 @@ async function prepareCodeImages(engine, images) {
 }
 
 // src/code-runtime-config.ts
-var CODE_PI_IMAGE = "ghcr.io/cory/odla-pi-agent@sha256:e7858b4f0291543171b2659b47c5d00f97acc4767cafd9b74ae5228899e0cde4";
+var CODE_PI_IMAGE = "ghcr.io/cory/odla-pi-agent@sha256:2a52a1c8eed66ad167e7bccab4d603ed8fcd75c42dddfb320083436ed89b7c9c";
 var CODE_BUILD_RECIPES = Object.freeze([{
   id: "odla-code-contracts",
   image: "node:24-alpine@sha256:a0b9bf06e4e6193cf7a0f58816cc935ff8c2a908f81e6f1a95432d679c54fbfd",
   command: [
     "node",
+    // The clean source mount deliberately has no node_modules. Reuse the
+    // semver copy inside this digest-pinned npm image without network access.
+    "--import=data:text/javascript,import%20Module%20from%20%22node%3Amodule%22%3Bprocess.env.NODE_PATH%3D%22%2Fusr%2Flocal%2Flib%2Fnode_modules%2Fnpm%2Fnode_modules%22%3BModule._initPaths%28%29",
     "--test",
     "scripts/lib/directories.test.mjs",
     "scripts/lib/internal-deps.test.mjs",