@odla-ai/cli 0.40.1 → 0.41.0

package/dist/bin.cjs

@@ -7915,7 +7915,7 @@ var init_code2 = __esm({
   }
 });
 
-// ../harness/dist/chunk-L2T3LPEF.js
+// ../harness/dist/chunk-5HX5LWTG.js
 async function digestStagedWorkspace(root, limits) {
   const files = [];
   const walk = async (directory) => {
@@ -9013,6 +9013,7 @@ async function runCodeAgent(options) {
 }
 async function runCodeAgentAttempt(options) {
   try {
+    const surface = options.surface ?? "v2";
     const { run } = await runCodeAgent({
       inference: options.inference,
       broker: options.broker,
@@ -9022,17 +9023,31 @@ 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);
@@ -9040,31 +9055,7 @@ 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,
@@ -9084,7 +9075,6 @@ 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);
@@ -9914,9 +9904,9 @@ async function appendCodeRuntimeEvent(control, command, event, refs) {
   const bounded = event.type === "message" ? { ...event, body: event.body.trim().slice(0, 2e4) || `${event.actor} event` } : event;
   await control.appendSessionEvent(command.sessionId, eventId, bounded);
 }
-var import_crypto, import_promises5, import_path5, import_child_process4, import_promises6, import_path6, import_child_process5, import_process2, import_crypto2, import_crypto3, import_fs2, import_promises7, import_path7, import_promises8, import_os3, import_path8, import_ai4, import_promises9, import_path9, import_promises10, import_promises11, import_path10, import_crypto4, CODE_RUNTIME_PROTOCOL_VERSION, CodeRuntimeReconciler, CodeRuntimeControlError, record5, invalid, RESERVED, SECRET, PATH, FORBIDDEN, ARTIFACT_PATH, PRIVATE_ARTIFACT_PART, SHA3, DIGEST3, ID3, RULE, DEFAULT_PREFIXES, DEFAULT_SUFFIXES, message, CodeRuntimeCheckpointManager, record22, SOURCE_LIMITS, RESERVED2, SECRET2, SOURCE_MAX_FILES, SOURCE_MAX_BYTES, SOURCE_SET_MAX_BYTES, V1_SYSTEM_PROMPT, V2_SYSTEM_PROMPT, V3_SYSTEM_PROMPT, SYSTEM_PROMPT_FOR, DEFAULT_MAX_FILES, DEFAULT_MAX_RESULTS, DEFAULT_MAX_FILE_BYTES, DESTINATIONS, READ, LIST, SEARCH, GRAPH, PATCH, RECIPE, cache, shortId, GRAPH_TOOLS, MAX_MEMORY_BODY, POSITIVE, digestRuntimeValue, runtimeErrorMessage, CodePiRuntimeEngine;
-var init_chunk_L2T3LPEF = __esm({
-  "../harness/dist/chunk-L2T3LPEF.js"() {
+var import_crypto, import_promises5, import_path5, import_child_process4, import_promises6, import_path6, import_child_process5, import_process2, import_crypto2, import_crypto3, import_fs2, import_promises7, import_path7, import_promises8, import_os3, import_path8, import_ai4, import_ai5, import_promises9, import_path9, import_promises10, import_promises11, import_path10, import_crypto4, CODE_RUNTIME_PROTOCOL_VERSION, CodeRuntimeReconciler, CodeRuntimeControlError, record5, invalid, RESERVED, SECRET, PATH, FORBIDDEN, ARTIFACT_PATH, PRIVATE_ARTIFACT_PART, SHA3, DIGEST3, ID3, RULE, DEFAULT_PREFIXES, DEFAULT_SUFFIXES, message, CodeRuntimeCheckpointManager, record22, SOURCE_LIMITS, RESERVED2, SECRET2, SOURCE_MAX_FILES, SOURCE_MAX_BYTES, SOURCE_SET_MAX_BYTES, V1_SYSTEM_PROMPT, V2_SYSTEM_PROMPT, V3_SYSTEM_PROMPT, SYSTEM_PROMPT_FOR, DEFAULT_MAX_FILES, DEFAULT_MAX_RESULTS, DEFAULT_MAX_FILE_BYTES, DESTINATIONS, READ, LIST, SEARCH, GRAPH, PATCH, RECIPE, cache, shortId, GRAPH_TOOLS, MAX_MEMORY_BODY, POSITIVE, digestRuntimeValue, runtimeErrorMessage, CodePiRuntimeEngine;
+var init_chunk_5HX5LWTG = __esm({
+  "../harness/dist/chunk-5HX5LWTG.js"() {
     "use strict";
     init_cjs_shims();
     init_chunk_K76I2TCQ();
@@ -9941,6 +9931,7 @@ var init_chunk_L2T3LPEF = __esm({
     import_os3 = require("os");
     import_path8 = require("path");
     import_ai4 = require("@odla-ai/ai");
+    import_ai5 = require("@odla-ai/ai");
     import_promises9 = require("fs/promises");
     import_path9 = require("path");
     init_dist();
@@ -10337,7 +10328,7 @@ Never claim a build or test passed unless odla_run_recipe returned that result.`
           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,
@@ -10352,31 +10343,30 @@ Never claim a build or test passed unless odla_run_recipe returned that result.`
           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 } : {}
         };
@@ -10438,7 +10428,7 @@ var init_node = __esm({
   "../harness/dist/node.js"() {
     "use strict";
     init_cjs_shims();
-    init_chunk_L2T3LPEF();
+    init_chunk_5HX5LWTG();
     init_chunk_K76I2TCQ();
     MEASURED_PREMIUM = Object.freeze({
       /** 3 racers vs pure depth at equal budget: 21,044 / 7,936. */
@@ -11519,12 +11509,12 @@ 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]
@@ -11992,6 +11982,7 @@ 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 ? `?${qs}` : ""}`);
   emit(ctx, page2, () => {
@@ -12098,11 +12089,16 @@ async function discussWho(ctx, parsed) {
   if (app) query.set("app", app);
   const kinds = stringOpt(parsed.options.kinds);
   if (kinds) query.set("kinds", kinds);
+  const topic = stringOpt(parsed.options.topic);
+  if (topic) query.set("topic", topic);
   const found = await request(ctx, "GET", `/mentionables?${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: ${found.failed.join(", ")})`);
   });
@@ -12457,6 +12453,8 @@ var init_discuss_command = __esm({
       "reopen",
       "by",
       "self",
+      "topic",
+      "mentions",
       "interval",
       "timeout",
       "cursor",
@@ -14752,7 +14750,7 @@ async function provision(options) {
       const key = import_node_process17.default.env[cfg.ai.keyEnv];
       if (key) {
         const secretName = cfg.ai.secretName ?? defaultSecretName(cfg.ai.provider);
-        await (0, import_ai5.putSecret)({ endpoint: cfg.dbEndpoint, token, fetch: doFetch }, tenantId, secretName, key);
+        await (0, import_ai6.putSecret)({ endpoint: cfg.dbEndpoint, token, fetch: doFetch }, tenantId, secretName, key);
         out.log(`${env}: ${cfg.ai.provider} key stored in vault (${secretName})`);
       } else {
         out.log(`${env}: ${cfg.ai.keyEnv} not set; skipped provider key storage`);
@@ -14788,13 +14786,13 @@ async function provision(options) {
     }
   }
 }
-var import_apps13, import_ai5, import_node_process17;
+var import_apps13, import_ai6, import_node_process17;
 var init_provision = __esm({
   "src/provision.ts"() {
     "use strict";
     init_cjs_shims();
     import_apps13 = require("@odla-ai/apps");
-    import_ai5 = require("@odla-ai/ai");
+    import_ai6 = require("@odla-ai/ai");
     import_node_process17 = __toESM(require("process"), 1);
     init_config();
     init_calendar();