@odla-ai/cli 0.40.1 → 0.41.0

package/dist/{cli-WJSQEBDO.js → cli-N5PASRYU.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   runCli
-} from "./chunk-XUTMQMXZ.js";
+} from "./chunk-6RGM4W73.js";
 import {
   exitCodeFor
 } from "./chunk-UKLSRQ5J.js";
@@ -9,4 +9,4 @@ export {
   exitCodeFor,
   runCli
 };
-//# sourceMappingURL=cli-WJSQEBDO.js.map
+//# sourceMappingURL=cli-N5PASRYU.js.map

package/dist/index.cjs

@@ -6275,7 +6275,7 @@ async function stageWorkspacePair(baselineSource, workspaceSource, options = {})
   }
 }
 
-// ../harness/dist/chunk-L2T3LPEF.js
+// ../harness/dist/chunk-5HX5LWTG.js
 var import_crypto = require("crypto");
 var import_promises5 = require("fs/promises");
 var import_path5 = require("path");
@@ -6612,7 +6612,7 @@ function validateSnapshot(snapshot, limits) {
   }
 }
 
-// ../harness/dist/chunk-L2T3LPEF.js
+// ../harness/dist/chunk-5HX5LWTG.js
 var import_child_process4 = require("child_process");
 var import_promises6 = require("fs/promises");
 var import_path6 = require("path");
@@ -6627,6 +6627,7 @@ var import_promises8 = require("fs/promises");
 var import_os3 = require("os");
 var import_path8 = require("path");
 var import_ai4 = require("@odla-ai/ai");
+var import_ai5 = require("@odla-ai/ai");
 var import_promises9 = require("fs/promises");
 var import_path9 = require("path");
 
@@ -6914,7 +6915,7 @@ function looksLikeDestination(value2) {
   return /^(?:[a-z][a-z0-9+.-]*:\/\/|\/|\\\\)/i.test(text3) || /^[\w.-]+\.[a-z]{2,}(?:[/:]|$)/i.test(text3);
 }
 
-// ../harness/dist/chunk-L2T3LPEF.js
+// ../harness/dist/chunk-5HX5LWTG.js
 var import_promises10 = require("fs/promises");
 var import_promises11 = require("fs/promises");
 var import_path10 = require("path");
@@ -7182,7 +7183,7 @@ async function buildCodeGraph(input) {
   return builder.build();
 }
 
-// ../harness/dist/chunk-L2T3LPEF.js
+// ../harness/dist/chunk-5HX5LWTG.js
 var import_crypto4 = require("crypto");
 async function digestStagedWorkspace(root, limits) {
   const files = [];
@@ -8428,6 +8429,7 @@ async function runCodeAgent(options) {
 }
 async function runCodeAgentAttempt(options) {
   try {
+    const surface = options.surface ?? "v2";
     const { run } = await runCodeAgent({
       inference: options.inference,
       broker: options.broker,
@@ -8437,17 +8439,31 @@ async function runCodeAgentAttempt(options) {
       // The brokered route resolves the real model from platform policy; this
       // id only labels the request the control plane is about to rewrite.
       model: "brokered",
-      surface: options.surface ?? "v2",
+      surface,
       ...options.maxSteps === void 0 ? {} : { maxSteps: options.maxSteps },
       ...options.budget ? { budget: options.budget } : {},
       ...options.signal ? { signal: options.signal } : {},
       ...options.onToolCall ? { onToolCall: options.onToolCall } : {}
     });
+    let finalText = run.finalText.trim();
+    if (!finalText && run.stoppedReason !== "refusal") {
+      const closing = await options.inference.chat({
+        model: "brokered",
+        system: `${SYSTEM_PROMPT_FOR[surface]}
+
+Finish with a concise, non-empty answer to the owner. Do not call tools or promise future work.`,
+        messages: [...run.messages, { role: "user", content: "Give the owner the closing answer now, grounded in the repository evidence and tool results above." }],
+        maxTokens: 16384,
+        ...options.signal ? { signal: options.signal } : {}
+      });
+      finalText = (0, import_ai5.extractText)(closing.content).trim();
+    }
+    const missingClosing = !finalText && run.stoppedReason !== "refusal";
     return {
-      status: run.stoppedReason === "refusal" ? "failed" : "completed",
-      finalText: run.finalText,
+      status: run.stoppedReason === "refusal" || missingClosing ? "failed" : "completed",
+      finalText,
       stoppedReason: run.stoppedReason,
-      ...run.stoppedReason === "refusal" ? { error: run.finalText || "the agent refused the task" } : {}
+      ...run.stoppedReason === "refusal" ? { error: finalText || "the agent refused the task" } : missingClosing ? { error: "the Code agent did not produce a closing answer" } : {}
     };
   } catch (cause) {
     const error = (cause instanceof Error ? cause.message : String(cause)).slice(0, 2e3);
@@ -8455,31 +8471,7 @@ async function runCodeAgentAttempt(options) {
   }
 }
 async function handleCodeRuntimeInference(input) {
-  const { command, metadata: metadata2, request: request3, state: state2 } = input;
-  if (state2.tokens >= metadata2.maxTokensPerInteraction) {
-    if (!state2.noticeEmitted) {
-      state2.noticeEmitted = true;
-      await input.event({
-        type: "message",
-        actor: "system",
-        body: `The agent paused at the ${metadata2.maxTokensPerInteraction.toLocaleString("en-US")}-token per-interaction limit. Send a new instruction to continue.`
-      }).catch(() => void 0);
-    }
-    return {
-      protocolVersion: HARNESS_PROTOCOL_VERSION,
-      type: "inference.response",
-      requestId: request3.requestId,
-      response: {
-        id: `budget:${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 { command, request: request3, state: state2 } = input;
   const startedAt = Date.now();
   const response2 = await input.control.infer(command.sessionId, {
     requestId: request3.requestId,
@@ -8499,7 +8491,6 @@ async function handleCodeRuntimeInference(input) {
     durationMs: Date.now() - startedAt,
     interactionId: command.commandId,
     interactionTokens: state2.tokens,
-    interactionMaxTokens: metadata2.maxTokensPerInteraction,
     ...costUsd === void 0 ? {} : { costUsd },
     ...state2.costKnown ? { interactionCostUsd: state2.costUsd } : {}
   }).catch(() => void 0);
@@ -9568,7 +9559,7 @@ var CodePiRuntimeEngine = class {
       recipeAuthorization: this.options.recipeAuthorization
     }, lease, metadata2.role));
     const startedAt = Date.now();
-    const interaction = { tokens: 0, noticeEmitted: false, costUsd: 0, costKnown: true };
+    const interaction = { tokens: 0, costUsd: 0, costKnown: true };
     const inference = createCodeRuntimeInference({
       command,
       metadata: metadata2,
@@ -9583,31 +9574,30 @@ var CodePiRuntimeEngine = class {
       lease,
       workspaceDir: active.workspace.workspaceDir,
       prompt: metadata2.prompt,
-      signal: active.abort.signal,
-      // The owner's per-interaction allowance, enforced by runAgent against
-      // INCREMENTAL usage. The control plane still reserves against the same
-      // ceiling, but this is what stops the loop cleanly at the boundary rather
-      // than letting it discover the limit through a synthesized pause reply.
-      budget: { maxTotalTokens: metadata2.maxTokensPerInteraction }
+      signal: active.abort.signal
     });
-    const body = result.finalText.trim() || (result.status === "completed" ? "The agent finished without a closing message." : result.error ?? "The agent failed.");
+    const closing = result.finalText.trim();
+    const completed = result.status === "completed" && Boolean(closing);
+    const detail = result.error?.trim() || (closing ? "the Code agent failed" : "the Code agent did not produce a closing answer");
+    const body = closing || detail;
     await this.#event(command, {
       type: "message",
-      actor: result.status === "completed" ? "agent" : "system",
+      actor: completed ? "agent" : "system",
       body
     }, active.conversationRefs).catch(() => void 0);
     await this.#event(command, {
       type: "status",
-      status: result.status === "completed" ? "idle" : "failed",
+      status: completed ? "idle" : "failed",
       durationMs: Date.now() - startedAt
     }, active.conversationRefs).catch(() => void 0);
-    if (result.status === "failed") {
-      const detail = (result.error ?? "").trim() || "the Code agent failed";
+    if (!completed) {
       await this.#diagnostic(command, active, detail);
       await this.#failure(command, active, detail);
     }
     return {
       ...result,
+      status: completed ? "completed" : "failed",
+      ...!completed ? { error: detail } : {},
       tokens: interaction.tokens,
       ...interaction.costKnown ? { costUsd: interaction.costUsd } : {}
     };
@@ -10694,12 +10684,12 @@ Usage:
   odla-ai pm <goal|task|decision|bug> rm <id>
   odla-ai pm handoff --app <id> [--project <id>] [--json]
   odla-ai discuss groups [--json]
-  odla-ai discuss list [--app <id>] [--q <text>] [--state open|resolved|all] [--json]
+  odla-ai discuss list [--app <id>] [--q <text>] [--state open|resolved|all] [--mentions] [--json]
   odla-ai discuss read <topic> [--limit <n> --offset <n>] [--json]
   odla-ai discuss post --app <id> --subject "..." --body "..." [--markup "... @[Label](kind/id)"] [--mutation-id <id>]
   odla-ai discuss reply <topic> --body "..." [--markup "..."] [--mutation-id <id>]
   odla-ai discuss resolve <topic> [--reopen] [--mutation-id <id>]
-  odla-ai discuss who --q <text> [--app <id>] [--kinds user,pm:task] [--json]
+  odla-ai discuss who --q <text> [--app <id>] [--topic <id>] [--kinds user,pm:task] [--json]
   odla-ai discuss watch [<topic>] [--cursor <cursor>] [--by <authorId>] [--self <authorId>] [--interval <s>] [--timeout <s>] [--json|--jsonl]
   odla-ai agent jobs [--env dev] [--state pending|running|succeeded|dead_letter] [--limit 50] [--token <ODLA_API_KEY>] [--json]
   odla-ai agent retry <job-id> [--env dev] [--token <ODLA_API_KEY>] [--json]
@@ -11138,6 +11128,7 @@ async function discussList(ctx, parsed) {
     const value2 = stringOpt(parsed.options[flag]);
     if (value2) query.set(param, value2);
   }
+  if (parsed.options.mentions === true) query.set("mentions", "1");
   const qs = query.toString();
   const page2 = await request(ctx, "GET", `/topics${qs ? `?${qs}` : ""}`);
   emit(ctx, page2, () => {
@@ -11244,11 +11235,16 @@ async function discussWho(ctx, parsed) {
   if (app) query.set("app", app);
   const kinds = stringOpt(parsed.options.kinds);
   if (kinds) query.set("kinds", kinds);
+  const topic = stringOpt(parsed.options.topic);
+  if (topic) query.set("topic", topic);
   const found = await request(ctx, "GET", `/mentionables?${query.toString()}`);
   emit(ctx, found, () => {
-    ctx.out.log("mention	kind	hint");
+    ctx.out.log("mention	kind	use	hint");
     for (const item of found.items) {
-      ctx.out.log(`@[${item.label}](${item.kind}/${item.id})	${item.kind}	${item.hint ?? ""}`);
+      const use = item.kind === "agent" ? item.meta?.executionAvailability === "runnable" ? "runnable" : "reference" : "";
+      ctx.out.log(
+        `@[${item.label}](${item.kind}/${item.id})	${item.kind}	${use}	${item.hint ?? ""}`
+      );
     }
     if (found.failed.length > 0) ctx.out.error(`(no results from: ${found.failed.join(", ")})`);
   });
@@ -11498,6 +11494,8 @@ var ALLOWED = [
   "reopen",
   "by",
   "self",
+  "topic",
+  "mentions",
   "interval",
   "timeout",
   "cursor",
@@ -13239,7 +13237,7 @@ function percent(value2) {
 
 // src/provision.ts
 var import_apps13 = require("@odla-ai/apps");
-var import_ai5 = require("@odla-ai/ai");
+var import_ai6 = require("@odla-ai/ai");
 var import_node_process17 = __toESM(require("process"), 1);
 
 // src/integration-provision.ts
@@ -13681,7 +13679,7 @@ async function provision(options) {
       const key = import_node_process17.default.env[cfg.ai.keyEnv];
       if (key) {
         const secretName = cfg.ai.secretName ?? defaultSecretName(cfg.ai.provider);
-        await (0, import_ai5.putSecret)({ endpoint: cfg.dbEndpoint, token, fetch: doFetch }, tenantId, secretName, key);
+        await (0, import_ai6.putSecret)({ endpoint: cfg.dbEndpoint, token, fetch: doFetch }, tenantId, secretName, key);
         out.log(`${env}: ${cfg.ai.provider} key stored in vault (${secretName})`);
       } else {
         out.log(`${env}: ${cfg.ai.keyEnv} not set; skipped provider key storage`);