negotium 0.1.20 → 0.1.21

package/dist/main.js

@@ -3707,7 +3707,7 @@ var init_claude_provider = __esm(async () => {
 });
 
 // ../../packages/core/src/version.ts
-var NEGOTIUM_VERSION = "0.1.20";
+var NEGOTIUM_VERSION = "0.1.21";
 
 // ../../packages/core/src/agents/codex-native-multi-agent.ts
 import { spawn as spawn3 } from "child_process";
@@ -6079,11 +6079,18 @@ __export(exports_model_catalog, {
   selectableModel: () => selectableModel,
   resolveModelForAgent: () => resolveModelForAgent,
   modelOwner: () => modelOwner,
+  formatSelectableModel: () => formatSelectableModel,
   SELECTABLE_MODELS: () => SELECTABLE_MODELS,
   MODEL_OWNER: () => MODEL_OWNER,
+  MODEL_COST_ROUTING_SUMMARY: () => MODEL_COST_ROUTING_SUMMARY,
+  MODEL_COST_RESEARCHED_AT: () => MODEL_COST_RESEARCHED_AT,
   FALLBACK_ORDER: () => FALLBACK_ORDER,
   AGENT_DISPLAY_NAME: () => AGENT_DISPLAY_NAME
 });
+function formatSelectableModel(candidate) {
+  const tier = `${candidate.intelligenceTier[0].toUpperCase()}${candidate.intelligenceTier.slice(1)}`;
+  return `${candidate.agent} / \`${candidate.model}\` [${tier}-level]: ${candidate.routingSummary}`;
+}
 function selectableModel(value) {
   const normalized = value.trim().toLowerCase();
   return SELECTABLE_MODELS.find((candidate) => candidate.model === normalized);
@@ -6106,7 +6113,7 @@ function resolveModelForAgent(agent, requested, registry) {
     return defaultModel;
   return registry.validateModel(requested) ? requested : defaultModel;
 }
-var MODEL_OWNER, SELECTABLE_MODELS, FALLBACK_ORDER, AGENT_DISPLAY_NAME;
+var MODEL_OWNER, MODEL_COST_RESEARCHED_AT = "2026-07-19", MODEL_COST_ROUTING_SUMMARY = "Cost basis (2026-07-19): Codex Pro 20x and Claude Max 20x are each $200/month; DeepSeek Pro is pay-per-token. Relative marginal token cost: DeepSeek Pro << Codex < Claude.", CODEX_PRO_20X_COST = "ChatGPT Pro 20x subscription: $200/month", CODEX_COMMUNITY_WEEKLY = "Community plan-level observation: roughly 2\u20134B raw/cached tokens per week; fresh-input equivalent is much lower and unstable (low confidence)", CLAUDE_MAX_20X_COST = "Claude Max 20x subscription: $200/month", CLAUDE_COMMUNITY_SESSION = "Community observations vary from roughly 220\u2013250K locally displayed tokens per 5-hour session to billions of cache-heavy raw tokens per week; calibrated reports value a full weekly allowance around $680\u2013$1,900 at API rates. Recent heavy-model reports reach the weekly cap after about 4\u20135 full sessions (low confidence; not a token cap)", SELECTABLE_MODELS, FALLBACK_ORDER, AGENT_DISPLAY_NAME;
 var init_model_catalog = __esm(() => {
   init_config();
   MODEL_OWNER = {
@@ -6126,47 +6133,72 @@ var init_model_catalog = __esm(() => {
     {
       model: "gpt-5.6-sol",
       agent: "codex",
-      description: "Latest frontier agentic coding model."
+      description: "Highest-capability Codex route for the hardest agentic coding work.",
+      intelligenceTier: "fable",
+      routingSummary: "hardest coding work; 5x Codex quota cost",
+      accessCost: CODEX_PRO_20X_COST,
+      marginalTokenCost: "Codex credits: $5/M uncached input, $0.50/M cached input, $30/M output",
+      estimatedUsage: `Official Pro 20x range: 300\u20131,800 local messages per 5 hours; quota weight 5x Luna. ${CODEX_COMMUNITY_WEEKLY}`
     },
     {
       model: "gpt-5.6-terra",
       agent: "codex",
-      description: "Balanced agentic coding model for everyday work."
+      description: "High-capability Codex route for complex coding and reasoning.",
+      intelligenceTier: "opus",
+      routingSummary: "complex coding and reasoning; 2.5x Codex quota cost",
+      accessCost: CODEX_PRO_20X_COST,
+      marginalTokenCost: "Codex credits: $2.50/M uncached input, $0.25/M cached input, $15/M output",
+      estimatedUsage: `Official Pro 20x range: 400\u20132,200 local messages per 5 hours; quota weight 2.5x Luna. ${CODEX_COMMUNITY_WEEKLY}`
     },
     {
       model: "gpt-5.6-luna",
       agent: "codex",
-      description: "Fast and affordable agentic coding model."
-    },
-    {
-      model: "gpt-5.5",
-      agent: "codex",
-      description: "Frontier model for complex coding, research, and real-world work."
+      description: "Default Codex route with strong everyday coding intelligence.",
+      intelligenceTier: "sonnet",
+      routingSummary: "everyday coding default; lowest Codex quota cost (1x)",
+      accessCost: CODEX_PRO_20X_COST,
+      marginalTokenCost: "Codex credits: $1/M uncached input, $0.10/M cached input, $6/M output",
+      estimatedUsage: `Official Pro 20x range: 1,000\u20135,600 local messages per 5 hours; lowest Codex quota weight (1x). ${CODEX_COMMUNITY_WEEKLY}`
     },
     {
       model: "fable",
       agent: "claude",
-      description: "Fable 5 \xB7 Most capable for your hardest and longest-running tasks \xB7 $10/$50 per Mtok"
+      description: "Highest-capability Claude route for the hardest and longest-running tasks.",
+      intelligenceTier: "fable",
+      routingSummary: "hardest long-running work; highest Claude cost; explicit request only",
+      accessCost: CLAUDE_MAX_20X_COST,
+      marginalTokenCost: "Claude API/extra usage: $10/M input, $12.50/M cache write, $1/M cache read, $50/M output",
+      estimatedUsage: `${CLAUDE_COMMUNITY_SESSION}; Fable drains weighted quota fastest, so use only on explicit user request. No stable per-model token cap is published.`
     },
     {
       model: "opus",
       agent: "claude",
-      description: "Opus 4.8 with 1M context \xB7 Best for everyday, complex tasks \xB7 $5/$25 per Mtok"
+      description: "High-capability Claude route for complex reasoning and tool-heavy work.",
+      intelligenceTier: "opus",
+      routingSummary: "complex reasoning and tool-heavy work; about 2.5x Sonnet marginal cost",
+      accessCost: CLAUDE_MAX_20X_COST,
+      marginalTokenCost: "Claude API/extra usage: $5/M input, $6.25/M cache write, $0.50/M cache read, $25/M output",
+      estimatedUsage: `${CLAUDE_COMMUNITY_SESSION}; Opus uses the shared all-model weekly pool more quickly than Sonnet. No stable per-model token cap is published.`
     },
     {
       model: "sonnet",
       agent: "claude",
-      description: "Sonnet 5 \xB7 Efficient for routine tasks \xB7 $2/$10 per Mtok \xB7 promo through Aug 31"
+      description: "Default Claude route for capable, efficient everyday work.",
+      intelligenceTier: "sonnet",
+      routingSummary: "capable everyday default; lowest Claude model cost",
+      accessCost: CLAUDE_MAX_20X_COST,
+      marginalTokenCost: "Claude API/extra usage introductory rate: $2/M input, $2.50/M cache write, $0.20/M cache read, $10/M output through 2026-08-31; then $3/M input and $15/M output",
+      estimatedUsage: `${CLAUDE_COMMUNITY_SESSION}; Sonnet also has a separate weekly allowance and normally provides the highest Claude throughput. No stable weekly token cap is published.`
     },
     {
       model: "deepseek-pro",
       agent: "maestro",
-      description: "Sonnet-level performance at lower cost \xB7 Best for everyday tasks over coding"
-    },
-    {
-      model: "deepseek-flash",
-      agent: "maestro",
-      description: "DeepSeek V4 Flash \xB7 Fast and low-cost for lightweight everyday tasks"
+      description: "API-priced Sonnet-level route for cost-efficient everyday work.",
+      intelligenceTier: "sonnet",
+      routingSummary: "cost-efficient everyday work; pay-per-token and cheapest route",
+      accessCost: "DeepSeek V4 Pro pay-as-you-go API; no monthly subscription required",
+      marginalTokenCost: "DeepSeek API: $0.435/M uncached input, $0.003625/M cached input, $0.87/M output",
+      estimatedUsage: "No subscription token cap; pay per token. Official account concurrency limit is 500 requests."
     }
   ];
   FALLBACK_ORDER = {
@@ -7120,6 +7152,7 @@ async function spawnPlaywright(instanceKey, userId, reservedPort, browserBin = P
     String(port),
     "--host",
     "127.0.0.1",
+    "--headed",
     "--user-data-dir",
     userDataDir
   ];
@@ -8446,7 +8479,7 @@ function createSelfConfigToolDefinitions(ctx) {
   return [
     {
       name: "set_model",
-      description: "Set the model for THIS topic. Persists and applies from the NEXT turn. Claude: 'sonnet' / 'opus' / 'fable'. Codex: 'gpt-5.6-luna' (default) / 'gpt-5.6-terra' / 'gpt-5.6-sol'. Maestro: 'deepseek-pro' / 'deepseek-flash' / 'deepseek'. The model must match the topic's current agent's accepted list. Switch agent first via set_agent only when crossing the claude/codex/maestro runtime boundary. Fails if the user locked this setting. NEVER use 'fable' unless the user explicitly requests it.",
+      description: "Set the model for THIS topic; it persists and applies from the NEXT turn. Use a model from the system prompt's same-agent catalog. Cross-agent changes require an explicit user-requested set_agent first. Fails if the user locked this setting. NEVER use 'fable' unless explicitly requested.",
       schema: { model: z2.string().describe("Model id valid for the topic's current agent.") },
       async handler({ model }) {
         return mcpResult(setSelfConfigModel(getCtx(), model));
@@ -8843,7 +8876,7 @@ function loadAgentPrompt(filename) {
     prompt: match[2].trim()
   };
 }
-function buildRuntimeToolSection(agentKind, canSpawnSubagents = false, visualTools = false, fileDeliveryTools = false) {
+function buildRuntimeToolSection(agentKind, canSpawnSubagents = false, visualTools = false, fileDeliveryTools = false, currentModel, currentEffort) {
   const runtimeNamespace = "mcp__runtime";
   const taskNamespace = "mcp__task";
   const visualToolLine = agentKind === "codex" ? `To display charts, tables, or interactive HTML results to the user, call the \`show_html\` function in the \`${runtimeNamespace}\` namespace with { html: "<complete HTML string>", title?: "optional title" }.` : `To display charts, tables, or interactive HTML results to the user, call the MCP tool "${runtimeNamespace}__show_html" with { html: "<complete HTML string>", title?: "optional title" }.`;
@@ -8901,20 +8934,26 @@ function buildRuntimeToolSection(agentKind, canSpawnSubagents = false, visualToo
     "Do not use session communication to make another topic perform destructive changes without the user's clear intent.",
     ...spawnSubagentSection
   ];
+  const modelCatalog = SELECTABLE_MODELS.map((candidate) => `- ${formatSelectableModel(candidate)}`);
   const topicConfig = [
     "",
     "## Topic Configuration (model / agent / effort)",
+    `Current execution: agent=\`${agentKind}\`, model=\`${currentModel ?? "unknown (call get_model)"}\`, effort=\`${currentEffort ?? "provider default/off"}\`.`,
     "The user's configured agent/model/effort is intentional. Preserve it by default.",
     `When the user explicitly asks to change the model, agent backend, or reasoning effort for THIS topic, call "${runtimeNamespace}__set_model", "${runtimeNamespace}__set_agent", or "${runtimeNamespace}__set_effort". The change applies from your NEXT turn. After calling, briefly confirm and the system will continue with the new setting.`,
     "`set_effort` is available but discouraged; use it only when the user explicitly requests an effort change.",
-    "`set_model` may be called autonomously only when the current model is clearly below the task's required capability, such as complex algorithm design, proof-level math, or broad multi-file refactoring. In that case, move up one step within the same agent and end the turn. Do not use vague task complexity as a trigger.",
+    "`set_model` may be called autonomously only when the current model is clearly below the task's required capability, such as complex algorithm design, proof-level math, or broad multi-file refactoring. Choose the best-fit model directly from the same-agent catalog; model selection is not a mandatory one-step ladder. End the turn after changing it. Do not use vague task complexity as a trigger.",
     "`set_agent` autonomous calls are forbidden. Only switch agent when the user explicitly asks to switch runtime, e.g. \u201Ccodex\uB85C \uAC00\u201D, \u201Cclaude\uB85C \uBC14\uAFD4\u201D.",
     "Never use `fable` unless the user explicitly requests it; it is expensive.",
     "",
-    "Accepted values:",
-    "- claude models: `sonnet` (default), `opus`, `fable`; efforts: `low`, `medium`, `high`, `xhigh`, `max`.",
-    "- codex models: `gpt-5.6-luna` (default), `gpt-5.6-terra`, `gpt-5.6-sol`; efforts: `low`, `medium`, `high`, `xhigh`, `max`.",
-    "- maestro models: `deepseek-pro` (default), `deepseek-flash`, `deepseek`; efforts: `low`, `medium`, `high`, `xhigh`, `max`.",
+    "Model catalog (capability/cost routing guidance):",
+    MODEL_COST_ROUTING_SUMMARY,
+    ...modelCatalog,
+    "",
+    "Accepted effort values:",
+    "- claude: `low`, `medium`, `high`, `xhigh`, `max`.",
+    "- codex: `low`, `medium`, `high`, `xhigh`, `max`.",
+    "- maestro: `low`, `medium`, `high`, `xhigh`, `max`.",
     "Agent guidance when the user explicitly asks to switch: `codex` for deepest reasoning and complex code/math; `claude` for tool-heavy MCP/file automation; `maestro` for inexpensive fast drafts and lighter experiments."
   ];
   if (agentKind !== "claude") {
@@ -8937,7 +8976,7 @@ function buildTopicSystemPrompt(opts) {
     TOPIC_TITLE: opts.topicTitle,
     WORKSPACE_CWD: opts.workspaceCwd,
     UPLOADS_DIR: uploadsDir
-  }) + buildRuntimeToolSection(opts.agentKind, opts.canSpawnSubagents, opts.visualTools, opts.fileDeliveryTools);
+  }) + buildRuntimeToolSection(opts.agentKind, opts.canSpawnSubagents, opts.visualTools, opts.fileDeliveryTools, opts.currentModel, opts.currentEffort);
   if (opts.description?.trim()) {
     prompt += `
 
@@ -8953,7 +8992,7 @@ function buildChannelSystemPrompt(opts) {
     TOPIC_TITLE: opts.topicTitle,
     WORKSPACE_CWD: opts.workspaceCwd,
     UPLOADS_DIR: uploadsDir
-  }) + buildRuntimeToolSection(opts.agentKind, false, opts.visualTools, opts.fileDeliveryTools);
+  }) + buildRuntimeToolSection(opts.agentKind, false, opts.visualTools, opts.fileDeliveryTools, opts.currentModel, opts.currentEffort);
 }
 function buildManagerSystemPrompt(opts) {
   return `${buildTopicSystemPrompt(opts)}
@@ -9003,6 +9042,7 @@ User-uploaded files for this Channel are copied under "{{UPLOADS_DIR}}" as attac
 This is the shared "General" hub of the user's workspace.
 Act as the workspace manager: orient the user across topics, summarize what is going on, and route focused work to the right room.`, _topicSystemPromptTemplate = null, _channelSystemPromptTemplate = null, _managerSystemPromptTemplate = null, _visualDesignGuide = null;
 var init_builders = __esm(() => {
+  init_model_catalog();
   init_config();
   init_logger();
   PROMPTS_DIR = resolve6(PROJECT_ROOT, "src/prompts");
@@ -9121,10 +9161,6 @@ function getArchiverDef() {
 }
 function runArchiverTurn(params) {
   const { userId, topicId, topicTitle, archivePath, messageCount, mode = "deleted-topic" } = params;
-  if (mode !== "active-topic" && messageCount < MIN_ARCHIVE_MESSAGES) {
-    logger.info({ userId, topicTitle, messageCount, min: MIN_ARCHIVE_MESSAGES }, "archiver: skipped \u2014 too few messages to distil");
-    return false;
-  }
   let archiverDef;
   try {
     archiverDef = getArchiverDef();
@@ -9332,7 +9368,7 @@ ${rolled.join(`
     logger.warn({ err: err2, topicTitle }, "archiver: failed to post #General notification");
   }
 }
-var MAX_BRIEF_ENTRIES = 8, MIN_ARCHIVE_MESSAGES = 4, MAX_SESSION_STEPS = 20, activeArchiverSessions, _archiverDef = null;
+var MAX_BRIEF_ENTRIES = 8, MAX_SESSION_STEPS = 20, activeArchiverSessions, _archiverDef = null;
 var init_archiver = __esm(async () => {
   await init_agents();
   await init_bus();
@@ -9347,6 +9383,26 @@ var init_archiver = __esm(async () => {
   activeArchiverSessions = new Map;
 });
 
+// ../../packages/core/src/agents/memory-archive-policy.ts
+function countMemoryArchiveExchanges(rows) {
+  let waitingForAssistant = false;
+  let completed = 0;
+  for (const row of rows) {
+    const assistant = row.author_id === "ai" || Boolean(row.agent_type);
+    const conversational = row.kind == null || row.kind === "message";
+    if (!assistant && row.author_id !== "system" && conversational) {
+      waitingForAssistant = true;
+      continue;
+    }
+    if (assistant && conversational && waitingForAssistant) {
+      completed++;
+      waitingForAssistant = false;
+    }
+  }
+  return completed;
+}
+var MIN_MEMORY_ARCHIVE_EXCHANGES = 6;
+
 // ../../packages/core/src/storage/topic-transcript.ts
 function parseJsonField(value) {
   if (!value)
@@ -9434,8 +9490,9 @@ function archiveTopicMessages(topicId, topicTitle, options = {}) {
       counter++;
     }
   }
-  logger.info({ topicId, topicTitle, archive: path, messageCount: rows.length, lastRowid }, "archiveTopicMessages: archived topic messages");
-  return { path, messageCount: rows.length, lastRowid };
+  const exchangeCount = countMemoryArchiveExchanges(rows);
+  logger.info({ topicId, topicTitle, archive: path, messageCount: rows.length, exchangeCount, lastRowid }, "archiveTopicMessages: archived topic messages");
+  return { path, messageCount: rows.length, exchangeCount, lastRowid };
 }
 var init_topic_archive = __esm(async () => {
   init_logger();
@@ -9653,6 +9710,7 @@ function archiveActiveTopicForMemory(topicId, userId, options) {
   if (!options.allowMentionOnly && topic.aiMode === "mention")
     return "mention-only-channel";
   let skipped = "empty";
+  let skipMemoryTurn = false;
   const claim = claimTopicArchiveJob(topicId, (afterRowid) => {
     const pending = getMessagesForTopicAfterRowid(topicId, afterRowid);
     const minMessages = options.minMessages;
@@ -9661,6 +9719,8 @@ function archiveActiveTopicForMemory(topicId, userId, options) {
       logger.debug({ topicId, topicTitle: topic.title, pending: pending.length, minMessages }, "topic-memory-archiver: skipped below threshold");
       return null;
     }
+    const exchangeCount = countMemoryArchiveExchanges(pending);
+    skipMemoryTurn = options.minExchanges !== undefined && exchangeCount < options.minExchanges;
     const archived = (options.archiveMessages ?? archiveTopicMessages)(topicId, topic.title, {
       afterRowid,
       reason: options.reason
@@ -9678,6 +9738,18 @@ function archiveActiveTopicForMemory(topicId, userId, options) {
   if (claim.kind === "busy")
     return "busy";
   const { job } = claim;
+  if (skipMemoryTurn) {
+    settleTopicArchiveJob(topicId, job.archivePath, true);
+    logger.info({
+      topicId,
+      topicTitle: topic.title,
+      messageCount: job.messageCount,
+      minExchanges: options.minExchanges,
+      archive: job.archivePath,
+      reason: options.reason
+    }, "topic-memory-archiver: raw snapshot preserved below exchange threshold");
+    return "below-threshold";
+  }
   const memoryTopic = getTopicMemoryOrigin(topicId) ?? topic;
   const launched = (options.launchArchiver ?? runArchiverTurn)({
     userId,
@@ -11618,6 +11690,7 @@ __export(exports_turn_runner, {
   isPathInside: () => isPathInside,
   isLikelyCurrentChannelTrigger: () => isLikelyCurrentChannelTrigger,
   ingestAttachment: () => ingestAttachment,
+  formatSelectableModel: () => formatSelectableModel,
   formatChannelTranscriptLine: () => formatChannelTranscriptLine,
   escapeRegExp: () => escapeRegExp2,
   escapeHtml: () => escapeHtml2,
@@ -11635,6 +11708,8 @@ __export(exports_turn_runner, {
   SESSION_EXPIRED_MSG: () => SESSION_EXPIRED_MSG,
   SELECTABLE_MODELS: () => SELECTABLE_MODELS,
   MODEL_OWNER: () => MODEL_OWNER,
+  MODEL_COST_ROUTING_SUMMARY: () => MODEL_COST_ROUTING_SUMMARY,
+  MODEL_COST_RESEARCHED_AT: () => MODEL_COST_RESEARCHED_AT,
   FALLBACK_ORDER: () => FALLBACK_ORDER,
   AGENT_DISPLAY_NAME: () => AGENT_DISPLAY_NAME
 });
@@ -12842,6 +12917,8 @@ function startAiTurn(params) {
     topicTitle: topic.title,
     workspaceCwd,
     agentKind,
+    currentModel: resolvedModel,
+    currentEffort: resolvedEffort,
     description: topic.description,
     canSpawnSubagents: peerBridge?.canSpawnSubagents ?? (topicRecord?.kind === "agent" && !topicRecord.isSubagent),
     visualTools,
@@ -13520,7 +13597,7 @@ async function deleteTopicCascadeImpl(topic, userId, options, deletingAncestorId
         logger.warn({ err: err2, topicId }, "deleteTopicCascade: archive failed but force=true - continuing with hard delete");
       }
     }
-    if (archived) {
+    if (archived && archived.exchangeCount >= MIN_MEMORY_ARCHIVE_EXCHANGES) {
       try {
         const memoryTopic = getTopicMemoryOrigin(topicId) ?? topic;
         runArchiverTurn({
@@ -14060,7 +14137,7 @@ function createSpawnSubagentToolDefinition(ctx) {
       task: z3.string().describe("Self-contained task brief for the subagent. It sees nothing else \u2014 include all context."),
       name: z3.string().optional().describe("Short name for the subagent room. Auto-generated when omitted."),
       agent: z3.enum(["claude", "codex", "maestro"]).optional().describe("Agent backend override. Defaults to this room's agent."),
-      model: z3.string().optional().describe("Model override. Defaults to the agent's default.")
+      model: z3.string().optional().describe(`Best-fit model override from the system prompt catalog. Omit agent+model to inherit ${ctx.agent}/${ctx.model ?? "default"}; overriding agent without model uses that agent's default.`)
     },
     async handler(input) {
       return spawnSubagent(ctx, input);
@@ -14568,6 +14645,7 @@ async function restartTopicSession(topicId, userId, reason = "topic-session-rest
     (options.archiveMemory ?? archiveActiveTopicForMemory)(topicId, userId, {
       reason: "reset",
       minMessages: 1,
+      minExchanges: MIN_MEMORY_ARCHIVE_EXCHANGES,
       allowMentionOnly: true,
       skipBusyCheck: true
     });
@@ -32811,4 +32889,4 @@ switch (command) {
   }
 }
 
-//# debugId=7766C8A27B2E5D3C64756E2164756E21
+//# debugId=C8F4C2CD3CFA25F564756E2164756E21