@markus-global/cli 0.9.3 → 0.9.4

package/dist/markus.mjs

@@ -5074,7 +5074,7 @@ function clampHeartbeatIntervalMs(ms) {
     return DEFAULT_HEARTBEAT_INTERVAL_MS;
   return Math.max(MIN_HEARTBEAT_INTERVAL_MS, Math.min(MAX_HEARTBEAT_INTERVAL_MS, Math.round(ms)));
 }
-var TASK_GET_NOTES_DEFAULT, TASK_GET_DELIVERABLES_DEFAULT, TASK_GET_STATUS_HISTORY_DEFAULT, REQUIREMENT_LIST_DEFAULT, ENTITY_COMMENTS_DEFAULT, PROMPT_TASK_NOTES_MAX, PROMPT_TASK_NOTE_CHARS, PROMPT_DEP_NOTES_MAX, PROMPT_DEP_NOTE_CHARS, PROMPT_DEP_DESC_CHARS, PROMPT_EXECUTION_ROUNDS_MAX, PROMPT_LOG_ENTRY_CHARS, PROMPT_RECENT_ERRORS_MAX, PROMPT_HUMAN_COMMENTS_MAX, REVIEWER_NOTES_MAX, SYSTEM_MY_TASKS_MAX, SYSTEM_TEAM_TASKS_MAX, SYSTEM_TASK_DESC_CHARS, PROMPT_PROJECT_DESC_CHARS, PROMPT_REQUIREMENT_DESC_CHARS, REVIEWER_FILE_LIST_MAX, DELIVERABLE_TITLE_CHARS, REVISION_REASON_CHARS, TASK_MAX_RETRIES, MAILBOX_ITEM_MAX_RETRIES, COMPLETION_MARKER, COMPLETION_MARKER_INSTRUCTION, COMPLETION_MARKER_LEAK_RE, TASK_MAX_NO_SUBMIT_RETRIES, TASK_RETRY_DELAYS_MS, PREEMPT_REQUEUE_DELAY_MS, SYSTEM_KNOWLEDGE_CHARS, SYSTEM_SOPS_CHARS, SYSTEM_LONGTERM_MEMORY_CHARS, MEMORY_MD_SECTION_MAX_CHARS, MEMORY_MD_TOTAL_MAX_CHARS, SYSTEM_LESSON_ENTRIES_MAX, SYSTEM_BEST_PRACTICE_ENTRIES_MAX, SYSTEM_DELIVERABLES_CHARS, SYSTEM_DAILY_LOG_CHARS, SYSTEM_DAILY_LOG_DAYS, SYSTEM_USER_PROFILE_CHARS, SYSTEM_CHAT_SESSIONS_MAX, HEARTBEAT_DAILY_LOG_CHARS, SYSTEM_PROJECT_DESC_CHARS, SYSTEM_DELIVERABLE_PREVIEW_CHARS, SYSTEM_MAILBOX_MERGED_CHARS, SYSTEM_MAILBOX_ITEM_PREVIEW_CHARS, TASK_LIST_PAGE_SIZE, TASK_LIST_PAGE_MAX, ARCHIVE_COMPLETED_AFTER_DAYS, ARCHIVE_TERMINAL_AFTER_DAYS, ARCHIVE_REQUIREMENT_AFTER_DAYS, ARCHIVE_SCAN_INTERVAL_MS, CHANNEL_CONTEXT_MESSAGES, MAILBOX_QUEUED_TTL_MS, MAILBOX_COALESCE_WINDOW_MS, MAILBOX_PROCESSING_TIMEOUT_MS, SSE_DISCONNECT_FORCE_STOP_MS, WATCHDOG_INTERVAL_MS, WATCHDOG_DRIFT_THRESHOLD_MS, TRIAGE_BACKLOG_THRESHOLD, TRIAGE_PROMPT_MAX_ITEMS, TRIAGE_MAX_TOKENS, TRIAGE_TEMPERATURE, TRIAGE_CONTEXT_MESSAGES_MAX, TRIAGE_CONTEXT_MSG_CHARS, TRIAGE_ITEM_CONTENT_CHARS, TRIAGE_STALE_INFO_TTL_MS, TRIAGE_STALE_DROP_TYPES, TRIAGE_MAX_TOOL_ITERATIONS, TRIAGE_ALLOWED_TOOLS, DELIBERATION_ALLOWED_TOOLS, DEFAULT_HEARTBEAT_INTERVAL_MS, MIN_HEARTBEAT_INTERVAL_MS, MAX_HEARTBEAT_INTERVAL_MS, NOTEBOOK_MAX_AGENT_ENTRIES, NOTEBOOK_MAX_CHARS_PER_ENTRY, WORKING_MEMORY_MAX_ENTRIES, WORKING_MEMORY_MAX_CHARS, SUBAGENT_TASK_PREVIEW_CHARS, SUBAGENT_THINKING_PREVIEW_CHARS, SUBAGENT_RESULT_PREVIEW_CHARS, SUBAGENT_LOG_ENTRY_CHARS, SUBAGENT_ERROR_PREVIEW_CHARS, SUBAGENT_MAX_PARALLEL, SUBAGENT_MAX_AGGREGATE_ITERATIONS, SUBAGENT_MAX_LLM_RETRIES, SUBAGENT_RETRY_BASE_MS, SHELL_TIMEOUT_DEFAULT_MS, SHELL_TIMEOUT_MAX_MS, SHELL_MAX_SESSIONS_PER_AGENT, SHELL_SESSION_IDLE_TIMEOUT_MS, SHELL_SESSION_MAX_OUTPUT_BYTES, APPROVAL_WAIT_TIMEOUT_MS, HEARTBEAT_MIN_INITIAL_DELAY_MS, HEARTBEAT_STARTUP_JITTER_MS, LLM_CIRCUIT_RESET_RATE_LIMIT_MS, SESSION_STORAGE_COMPACT_TRIGGER, SESSION_STORAGE_COMPACT_KEEP, SESSION_STORAGE_TOOL_SHRINK_CHARS, CONTEXT_ABSURD_MESSAGE_CHARS, CONTEXT_PROACTIVE_COMPACT_RATIO, SESSION_RESTORE_MAX_MESSAGES, SESSION_RESTORE_MAX_MESSAGE_TOKENS, TOOL_RESULT_OFFLOAD_CHARS, SYSTEM_COLLEAGUES_MAX, SYSTEM_OTHER_TEAMS_MAX, SYSTEM_OTHER_TEAM_MEMBERS_MAX, SYSTEM_HUMANS_MAX, PROMPT_AFFORD_OUTPUT_RESERVE, TOOL_DEF_BUDGET_REFLEX, TOOL_DEF_BUDGET_CONVERSE, TOOL_DEF_BUDGET_EXECUTE, TOOL_DEF_BUDGET_GOVERN, ROLE_PROMPT_MAX_TOKENS, KNOWLEDGE_PROMPT_MAX_TOKENS, KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX, STATE_PROMPT_MAX_LINES_REFLEX, STATE_TTL_DAYS, COLD_CONVERSE_FIXED_MAX, COLD_REFLEX_FIXED_MAX, SYSTEM_PROMPT_BUDGET_CONVERSE, DEFERRED_CATALOG_MAX_CHARS, MAX_TOKENS_AFFORD_SAFETY, MAX_TOKENS_REMAINING_MARGIN, MAX_TOKENS_CLAMP_FLOOR, DEEP_SLEEP_IDLE_HEARTBEATS, SUBTASK_SOFT_CAP, PROMPT_AFFORD_SAFETY_MARGIN, LLM_MAX_CONCURRENT_PER_PROVIDER, LLM_CONCURRENCY_JITTER_BASE_MS;
+var TASK_GET_NOTES_DEFAULT, TASK_GET_DELIVERABLES_DEFAULT, TASK_GET_STATUS_HISTORY_DEFAULT, REQUIREMENT_LIST_DEFAULT, ENTITY_COMMENTS_DEFAULT, PROMPT_TASK_NOTES_MAX, PROMPT_TASK_NOTE_CHARS, PROMPT_DEP_NOTES_MAX, PROMPT_DEP_NOTE_CHARS, PROMPT_DEP_DESC_CHARS, PROMPT_EXECUTION_ROUNDS_MAX, PROMPT_LOG_ENTRY_CHARS, PROMPT_RECENT_ERRORS_MAX, PROMPT_HUMAN_COMMENTS_MAX, REVIEWER_NOTES_MAX, SYSTEM_MY_TASKS_MAX, SYSTEM_TEAM_TASKS_MAX, SYSTEM_TASK_DESC_CHARS, PROMPT_PROJECT_DESC_CHARS, PROMPT_REQUIREMENT_DESC_CHARS, REVIEWER_FILE_LIST_MAX, DELIVERABLE_TITLE_CHARS, REVISION_REASON_CHARS, TASK_MAX_RETRIES, MAILBOX_ITEM_MAX_RETRIES, COMPLETION_MARKER, COMPLETION_MARKER_INSTRUCTION, COMPLETION_MARKER_LEAK_RE, TASK_MAX_NO_SUBMIT_RETRIES, TASK_RETRY_DELAYS_MS, PREEMPT_REQUEUE_DELAY_MS, SYSTEM_KNOWLEDGE_CHARS, SYSTEM_SOPS_CHARS, SYSTEM_LONGTERM_MEMORY_CHARS, MEMORY_MD_SECTION_MAX_CHARS, MEMORY_MD_TOTAL_MAX_CHARS, SYSTEM_LESSON_ENTRIES_MAX, SYSTEM_BEST_PRACTICE_ENTRIES_MAX, SYSTEM_DELIVERABLES_CHARS, SYSTEM_DAILY_LOG_CHARS, SYSTEM_DAILY_LOG_DAYS, SYSTEM_USER_PROFILE_CHARS, SYSTEM_CHAT_SESSIONS_MAX, HEARTBEAT_DAILY_LOG_CHARS, SYSTEM_PROJECT_DESC_CHARS, SYSTEM_DELIVERABLE_PREVIEW_CHARS, SYSTEM_MAILBOX_MERGED_CHARS, SYSTEM_MAILBOX_ITEM_PREVIEW_CHARS, TASK_LIST_PAGE_SIZE, TASK_LIST_PAGE_MAX, ARCHIVE_COMPLETED_AFTER_DAYS, ARCHIVE_TERMINAL_AFTER_DAYS, ARCHIVE_REQUIREMENT_AFTER_DAYS, ARCHIVE_SCAN_INTERVAL_MS, CHANNEL_CONTEXT_MESSAGES, MAILBOX_QUEUED_TTL_MS, MAILBOX_COALESCE_WINDOW_MS, MAILBOX_PROCESSING_TIMEOUT_MS, SSE_DISCONNECT_FORCE_STOP_MS, WATCHDOG_INTERVAL_MS, WATCHDOG_DRIFT_THRESHOLD_MS, TRIAGE_BACKLOG_THRESHOLD, TRIAGE_PROMPT_MAX_ITEMS, TRIAGE_MAX_TOKENS, TRIAGE_TEMPERATURE, TRIAGE_CONTEXT_MESSAGES_MAX, TRIAGE_CONTEXT_MSG_CHARS, TRIAGE_ITEM_CONTENT_CHARS, TRIAGE_STALE_INFO_TTL_MS, TRIAGE_STALE_DROP_TYPES, TRIAGE_MAX_TOOL_ITERATIONS, TRIAGE_ALLOWED_TOOLS, DELIBERATION_ALLOWED_TOOLS, DEFAULT_HEARTBEAT_INTERVAL_MS, MIN_HEARTBEAT_INTERVAL_MS, MAX_HEARTBEAT_INTERVAL_MS, NOTEBOOK_MAX_AGENT_ENTRIES, NOTEBOOK_MAX_CHARS_PER_ENTRY, WORKING_MEMORY_MAX_ENTRIES, WORKING_MEMORY_MAX_CHARS, SUBAGENT_TASK_PREVIEW_CHARS, SUBAGENT_THINKING_PREVIEW_CHARS, SUBAGENT_RESULT_PREVIEW_CHARS, SUBAGENT_LOG_ENTRY_CHARS, SUBAGENT_ERROR_PREVIEW_CHARS, SUBAGENT_MAX_PARALLEL, SUBAGENT_MAX_AGGREGATE_ITERATIONS, SUBAGENT_MAX_LLM_RETRIES, SUBAGENT_RETRY_BASE_MS, SHELL_TIMEOUT_DEFAULT_MS, SHELL_TIMEOUT_MAX_MS, SHELL_MAX_SESSIONS_PER_AGENT, SHELL_SESSION_IDLE_TIMEOUT_MS, SHELL_SESSION_MAX_OUTPUT_BYTES, APPROVAL_WAIT_TIMEOUT_MS, HEARTBEAT_MIN_INITIAL_DELAY_MS, HEARTBEAT_STARTUP_JITTER_MS, LLM_CIRCUIT_RESET_RATE_LIMIT_MS, SESSION_STORAGE_COMPACT_TRIGGER, SESSION_STORAGE_COMPACT_KEEP, SESSION_STORAGE_TOOL_SHRINK_CHARS, CONTEXT_ABSURD_MESSAGE_CHARS, CONTEXT_PROACTIVE_COMPACT_RATIO, SESSION_RESTORE_MAX_MESSAGES, SESSION_RESTORE_MAX_MESSAGE_TOKENS, TOOL_RESULT_OFFLOAD_CHARS, SYSTEM_COLLEAGUES_MAX, SYSTEM_OTHER_TEAMS_MAX, SYSTEM_OTHER_TEAM_MEMBERS_MAX, SYSTEM_HUMANS_MAX, PROMPT_AFFORD_OUTPUT_RESERVE, TOOL_DEF_BUDGET_REFLEX, TOOL_DEF_BUDGET_CONVERSE, TOOL_DEF_BUDGET_EXECUTE, TOOL_DEF_BUDGET_GOVERN, ROLE_PROMPT_MAX_TOKENS, KNOWLEDGE_PROMPT_MAX_TOKENS, KNOWLEDGE_PROMPT_MAX_TOKENS_CONVERSE, KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX, STATE_PROMPT_MAX_LINES_REFLEX, STATE_TTL_DAYS, COLD_CONVERSE_FIXED_MAX, COLD_REFLEX_FIXED_MAX, SYSTEM_PROMPT_BUDGET_CONVERSE, SYSTEM_ANNOUNCEMENTS_CHARS_CONVERSE, SYSTEM_NORMS_CHARS_CONVERSE, SYSTEM_ANNOUNCEMENTS_CHARS, SYSTEM_NORMS_CHARS, SYSTEM_WORKFLOWS_MAX_CONVERSE, SYSTEM_WORKFLOW_DESC_CHARS_CONVERSE, SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE, DEFERRED_CATALOG_MAX_CHARS, MAX_TOKENS_AFFORD_SAFETY, MAX_TOKENS_REMAINING_MARGIN, MAX_TOKENS_CLAMP_FLOOR, DEEP_SLEEP_IDLE_HEARTBEATS, SUBTASK_SOFT_CAP, PROMPT_AFFORD_SAFETY_MARGIN, LLM_MAX_CONCURRENT_PER_PROVIDER, LLM_CONCURRENCY_JITTER_BASE_MS;
 var init_limits = __esm({
   "../shared/dist/limits.js"() {
     "use strict";
@@ -5247,12 +5247,20 @@ var init_limits = __esm({
     TOOL_DEF_BUDGET_GOVERN = 8e3;
     ROLE_PROMPT_MAX_TOKENS = 2500;
     KNOWLEDGE_PROMPT_MAX_TOKENS = 1500;
+    KNOWLEDGE_PROMPT_MAX_TOKENS_CONVERSE = 1e3;
     KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX = 0;
     STATE_PROMPT_MAX_LINES_REFLEX = 5;
     STATE_TTL_DAYS = 7;
     COLD_CONVERSE_FIXED_MAX = 12e3;
     COLD_REFLEX_FIXED_MAX = 8e3;
     SYSTEM_PROMPT_BUDGET_CONVERSE = 8e3;
+    SYSTEM_ANNOUNCEMENTS_CHARS_CONVERSE = 400;
+    SYSTEM_NORMS_CHARS_CONVERSE = 400;
+    SYSTEM_ANNOUNCEMENTS_CHARS = 2e3;
+    SYSTEM_NORMS_CHARS = 2e3;
+    SYSTEM_WORKFLOWS_MAX_CONVERSE = 3;
+    SYSTEM_WORKFLOW_DESC_CHARS_CONVERSE = 80;
+    SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE = 800;
     DEFERRED_CATALOG_MAX_CHARS = 1500;
     MAX_TOKENS_AFFORD_SAFETY = 64;
     MAX_TOKENS_REMAINING_MARGIN = 500;
@@ -5379,6 +5387,7 @@ __export(dist_exports, {
   HEARTBEAT_MIN_INITIAL_DELAY_MS: () => HEARTBEAT_MIN_INITIAL_DELAY_MS,
   HEARTBEAT_STARTUP_JITTER_MS: () => HEARTBEAT_STARTUP_JITTER_MS,
   KNOWLEDGE_PROMPT_MAX_TOKENS: () => KNOWLEDGE_PROMPT_MAX_TOKENS,
+  KNOWLEDGE_PROMPT_MAX_TOKENS_CONVERSE: () => KNOWLEDGE_PROMPT_MAX_TOKENS_CONVERSE,
   KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX: () => KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX,
   LLM_CIRCUIT_RESET_RATE_LIMIT_MS: () => LLM_CIRCUIT_RESET_RATE_LIMIT_MS,
   LLM_CONCURRENCY_JITTER_BASE_MS: () => LLM_CONCURRENCY_JITTER_BASE_MS,
@@ -5447,6 +5456,8 @@ __export(dist_exports, {
   SUBAGENT_TASK_PREVIEW_CHARS: () => SUBAGENT_TASK_PREVIEW_CHARS,
   SUBAGENT_THINKING_PREVIEW_CHARS: () => SUBAGENT_THINKING_PREVIEW_CHARS,
   SUBTASK_SOFT_CAP: () => SUBTASK_SOFT_CAP,
+  SYSTEM_ANNOUNCEMENTS_CHARS: () => SYSTEM_ANNOUNCEMENTS_CHARS,
+  SYSTEM_ANNOUNCEMENTS_CHARS_CONVERSE: () => SYSTEM_ANNOUNCEMENTS_CHARS_CONVERSE,
   SYSTEM_BEST_PRACTICE_ENTRIES_MAX: () => SYSTEM_BEST_PRACTICE_ENTRIES_MAX,
   SYSTEM_CHAT_SESSIONS_MAX: () => SYSTEM_CHAT_SESSIONS_MAX,
   SYSTEM_COLLEAGUES_MAX: () => SYSTEM_COLLEAGUES_MAX,
@@ -5454,6 +5465,7 @@ __export(dist_exports, {
   SYSTEM_DAILY_LOG_DAYS: () => SYSTEM_DAILY_LOG_DAYS,
   SYSTEM_DELIVERABLES_CHARS: () => SYSTEM_DELIVERABLES_CHARS,
   SYSTEM_DELIVERABLE_PREVIEW_CHARS: () => SYSTEM_DELIVERABLE_PREVIEW_CHARS,
+  SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE: () => SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE,
   SYSTEM_HUMANS_MAX: () => SYSTEM_HUMANS_MAX,
   SYSTEM_KNOWLEDGE_CHARS: () => SYSTEM_KNOWLEDGE_CHARS,
   SYSTEM_LESSON_ENTRIES_MAX: () => SYSTEM_LESSON_ENTRIES_MAX,
@@ -5461,6 +5473,8 @@ __export(dist_exports, {
   SYSTEM_MAILBOX_ITEM_PREVIEW_CHARS: () => SYSTEM_MAILBOX_ITEM_PREVIEW_CHARS,
   SYSTEM_MAILBOX_MERGED_CHARS: () => SYSTEM_MAILBOX_MERGED_CHARS,
   SYSTEM_MY_TASKS_MAX: () => SYSTEM_MY_TASKS_MAX,
+  SYSTEM_NORMS_CHARS: () => SYSTEM_NORMS_CHARS,
+  SYSTEM_NORMS_CHARS_CONVERSE: () => SYSTEM_NORMS_CHARS_CONVERSE,
   SYSTEM_OTHER_TEAMS_MAX: () => SYSTEM_OTHER_TEAMS_MAX,
   SYSTEM_OTHER_TEAM_MEMBERS_MAX: () => SYSTEM_OTHER_TEAM_MEMBERS_MAX,
   SYSTEM_PROJECT_DESC_CHARS: () => SYSTEM_PROJECT_DESC_CHARS,
@@ -5469,6 +5483,8 @@ __export(dist_exports, {
   SYSTEM_TASK_DESC_CHARS: () => SYSTEM_TASK_DESC_CHARS,
   SYSTEM_TEAM_TASKS_MAX: () => SYSTEM_TEAM_TASKS_MAX,
   SYSTEM_USER_PROFILE_CHARS: () => SYSTEM_USER_PROFILE_CHARS,
+  SYSTEM_WORKFLOWS_MAX_CONVERSE: () => SYSTEM_WORKFLOWS_MAX_CONVERSE,
+  SYSTEM_WORKFLOW_DESC_CHARS_CONVERSE: () => SYSTEM_WORKFLOW_DESC_CHARS_CONVERSE,
   TASK_GET_DELIVERABLES_DEFAULT: () => TASK_GET_DELIVERABLES_DEFAULT,
   TASK_GET_NOTES_DEFAULT: () => TASK_GET_NOTES_DEFAULT,
   TASK_GET_STATUS_HISTORY_DEFAULT: () => TASK_GET_STATUS_HISTORY_DEFAULT,
@@ -8401,6 +8417,8 @@ var init_context_engine = __esm({
         const isDream = opts.scenario === "memory_consolidation";
         const promptProfile = opts.promptProfile ?? packToPromptProfile(scenarioToPack(opts.scenario));
         const isReflex = promptProfile === "reflex";
+        const isConverse = promptProfile === "converse";
+        const isExecuteLike = promptProfile === "execute" || promptProfile === "govern";
         const stable = [];
         {
           const roleText = opts.role.systemPrompt ?? "";
@@ -8422,23 +8440,29 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
           stable.push("\n## Tool Usage Rules");
           stable.push("**File editing discipline**: You MUST use `file_write` and `file_edit` for all file creation and modification. NEVER use `shell_execute` with `cat`, `echo`, `printf`, `tee`, pipes (`|`), output redirection (`>`, `>>`), heredocs (`<<`), or `sed`/`awk` to write or modify files \u2014 these bypass file access controls. `shell_execute` is for running commands (build, test, git, etc.), not for writing files.");
           stable.push("**Large file writing**: NEVER write a document >200 lines in a single `file_write` call. Write section by section: `file_write` the first section, then `file_edit` to append each subsequent section.");
-          stable.push("**Subagent delegation**: For heavy subtasks needing many tool calls or lots of file reading, delegate to `spawn_subagent` to keep your context lean. Use `spawn_subagents` to run independent subtasks in parallel.");
+          if (isExecuteLike) {
+            stable.push("**Subagent delegation**: For heavy subtasks needing many tool calls or lots of file reading, delegate to `spawn_subagent` to keep your context lean. Use `spawn_subagents` to run independent subtasks in parallel.");
+          }
           stable.push("**Built-in tools over CLI**: ALWAYS use built-in tools (`task_create`, `task_assign`, `package_install`, `agent_send_message`, `memory_save`, etc.) \u2014 NEVER run `markus` CLI commands via `shell_execute`. The CLI is strictly for human operators (server start, emergency stop, initial setup). Agents must use their native tool interface for all operations.");
           stable.push('**No auto-install/deploy (agents/teams)**: NEVER automatically hire/deploy agents or teams via `package_install` or `hub_install` unless explicitly requested by a human (e.g., "install", "deploy", "hire", "start"). Creating builder-artifacts is separate from deploying. **Skills** follow Learning Habits impact rules below (low-impact may install directly).');
           stable.push("");
           stable.push("\n## Search & Exploration Strategy");
-          stable.push("When you need to understand code or find information, use a layered approach \u2014 each layer is a fallback for the previous:");
-          stable.push("1. **Semantic search** (`memory_search`, `deliverable_search`): Start with conceptual queries to find relevant knowledge and existing outputs.");
-          stable.push("2. **Pattern search** (`grep_search`): Use for exact symbol names, error messages, configuration keys, or specific strings.");
-          stable.push("3. **File browsing** (`file_read`, `list_directory`): Navigate directory structure and read specific files when you know the likely location.");
-          stable.push("4. **External research** (`web_search`, `web_fetch`): Use for unfamiliar libraries, APIs, error codes, or best practices not found in the codebase.");
-          const hasBrowserSkill = opts.availableSkills?.some((s2) => s2.name === "chrome-devtools");
-          if (hasBrowserSkill) {
-            stable.push("5. **Browser tools** (`browser_navigate`, `browser_snapshot`, `browser_click`): when `web_search`/`web_fetch` fails (network error, JS-rendered page, rate-limiting), access the page interactively \u2014 handles JS rendering, auth flows, and complex navigation `web_fetch` cannot.");
+          if (isExecuteLike) {
+            stable.push("When you need to understand code or find information, use a layered approach \u2014 each layer is a fallback for the previous:");
+            stable.push("1. **Semantic search** (`memory_search`, `deliverable_search`): Start with conceptual queries to find relevant knowledge and existing outputs.");
+            stable.push("2. **Pattern search** (`grep_search`): Use for exact symbol names, error messages, configuration keys, or specific strings.");
+            stable.push("3. **File browsing** (`file_read`, `list_directory`): Navigate directory structure and read specific files when you know the likely location.");
+            stable.push("4. **External research** (`web_search`, `web_fetch`): Use for unfamiliar libraries, APIs, error codes, or best practices not found in the codebase.");
+            const hasBrowserSkill = opts.availableSkills?.some((s2) => s2.name === "chrome-devtools");
+            if (hasBrowserSkill) {
+              stable.push("5. **Browser tools** (`browser_navigate`, `browser_snapshot`, `browser_click`): when `web_search`/`web_fetch` fails (network error, JS-rendered page, rate-limiting), access the page interactively \u2014 handles JS rendering, auth flows, and complex navigation `web_fetch` cannot.");
+            } else {
+              stable.push("If `web_search`/`web_fetch` fails, try alternative queries or URLs, or `web_fetch` a search-engine URL directly (e.g. `https://www.google.com/search?q=YOUR_QUERY`). The `chrome-devtools` skill adds browser tools for JS-rendered/interactive sites.");
+            }
+            stable.push("Always check existing patterns in the codebase before introducing new conventions. When exploring unfamiliar code, start from entry points and trace data flow.");
           } else {
-            stable.push("If `web_search`/`web_fetch` fails, try alternative queries or URLs, or `web_fetch` a search-engine URL directly (e.g. `https://www.google.com/search?q=YOUR_QUERY`). The `chrome-devtools` skill adds browser tools for JS-rendered/interactive sites.");
+            stable.push("Layered lookup: semantic (`memory_search` / `deliverable_search`) \u2192 exact (`grep_search`) \u2192 files (`file_read` / `list_directory`) \u2192 web (`web_search` / `web_fetch`). Prefer existing codebase patterns. Activate `chrome-devtools` via `discover_tools` when pages need a real browser.");
           }
-          stable.push("Always check existing patterns in the codebase before introducing new conventions. When exploring unfamiliar code, start from entry points and trace data flow.");
           stable.push("");
           stable.push("\n## Learning Habits");
           stable.push("Get smarter over time. Prefer the lightest store that changes future behavior.");
@@ -8498,10 +8522,20 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
         if (orgCtx)
           semiStable.push(orgCtx);
         if (opts.teamAnnouncements?.trim()) {
-          semiStable.push("\n## Team Announcements\n" + opts.teamAnnouncements.trim());
+          const raw = opts.teamAnnouncements.trim();
+          const cap = isConverse ? SYSTEM_ANNOUNCEMENTS_CHARS_CONVERSE : SYSTEM_ANNOUNCEMENTS_CHARS;
+          const body = raw.length > cap ? `${raw.slice(0, cap)}
+
+_[announcements truncated \u2014 read \`${opts.teamDataDir ?? "team"}/ANNOUNCEMENT.md\` for full text]_` : raw;
+          semiStable.push("\n## Team Announcements\n" + body);
         }
         if (opts.teamNorms?.trim()) {
-          semiStable.push("\n## Team Working Norms\n" + opts.teamNorms.trim());
+          const raw = opts.teamNorms.trim();
+          const cap = isConverse ? SYSTEM_NORMS_CHARS_CONVERSE : SYSTEM_NORMS_CHARS;
+          const body = raw.length > cap ? `${raw.slice(0, cap)}
+
+_[norms truncated \u2014 read \`${opts.teamDataDir ?? "team"}/NORMS.md\` for full text]_` : raw;
+          semiStable.push("\n## Team Working Norms\n" + body);
         }
         if (opts.teamDataDir) {
           const lines = ["\n## Team Data Directory", `Path: \`${opts.teamDataDir}\``, "Files:", "- `ANNOUNCEMENT.md` \u2014 team announcements", "- `NORMS.md` \u2014 team working norms"];
@@ -8576,7 +8610,7 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
         if (opts.environment) {
           semiStable.push(this.buildEnvironmentSection(opts.environment));
         }
-        const knowledgeTokCap = isReflex ? KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX : KNOWLEDGE_PROMPT_MAX_TOKENS;
+        const knowledgeTokCap = isReflex ? KNOWLEDGE_PROMPT_MAX_TOKENS_REFLEX : isConverse ? KNOWLEDGE_PROMPT_MAX_TOKENS_CONVERSE : KNOWLEDGE_PROMPT_MAX_TOKENS;
         if (knowledgeTokCap > 0) {
           const longTermMem = opts.memory.getLongTermMemory();
           if (longTermMem) {
@@ -8598,9 +8632,11 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
         }
         const scenario = opts.scenario ?? "chat";
         semiStable.push(this.buildScenarioSection(scenario, { a2aWaitForReply: opts.a2aWaitForReply, isManager: opts.isTeamManager, channelKey: opts.channelKey }));
-        const scenarioPolicies = this.buildScenarioPolicyBlocks(scenario);
-        if (scenarioPolicies)
-          semiStable.push(scenarioPolicies);
+        if (isExecuteLike) {
+          const scenarioPolicies = this.buildScenarioPolicyBlocks(scenario);
+          if (scenarioPolicies)
+            semiStable.push(scenarioPolicies);
+        }
         const dynamic = [];
         if (opts.projectContext) {
           const { project, repositories, governanceRules, teamRole } = opts.projectContext;
@@ -8680,7 +8716,7 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
                 dynamic.push(`_(${otherDone.length} other completed/closed tasks)_`);
               }
             }
-          } else {
+          } else if (isExecuteLike) {
             dynamic.push("\n## Task Board");
             dynamic.push("No tasks on the board.");
           }
@@ -8696,14 +8732,26 @@ _[ROLE truncated to ${ROLE_PROMPT_MAX_TOKENS} tok budget]_` : roleText);
           }
           if (wc.availableWorkflows.length > 0) {
             dynamic.push("\n## Available Workflows");
-            for (const w of wc.availableWorkflows) {
-              dynamic.push(`- **${w.name}** (${w.stepCount} steps): ${w.description}`);
+            const list = isConverse ? wc.availableWorkflows.slice(0, SYSTEM_WORKFLOWS_MAX_CONVERSE) : wc.availableWorkflows;
+            for (const w of list) {
+              const desc = isConverse ? w.description.slice(0, SYSTEM_WORKFLOW_DESC_CHARS_CONVERSE) : w.description;
+              dynamic.push(`- **${w.name}** (${w.stepCount} steps): ${desc}`);
+            }
+            if (isConverse && wc.availableWorkflows.length > SYSTEM_WORKFLOWS_MAX_CONVERSE) {
+              dynamic.push(`_(${wc.availableWorkflows.length - SYSTEM_WORKFLOWS_MAX_CONVERSE} more \u2014 use \`workflow_list\`)_`);
             }
             dynamic.push("Use `workflow_list` for details or `workflow_run` to start a workflow.");
           }
         }
         if (opts.dynamicContext) {
-          dynamic.push(opts.dynamicContext);
+          const raw = opts.dynamicContext;
+          if (isConverse && raw.length > SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE) {
+            dynamic.push(`${raw.slice(0, SYSTEM_DYNAMIC_CONTEXT_CHARS_CONVERSE)}
+
+_[dynamic context truncated]_`);
+          } else {
+            dynamic.push(raw);
+          }
         }
         const alreadyShownIds = /* @__PURE__ */ new Set();
         const cpp = opts.cognitiveContext;
@@ -8843,14 +8891,35 @@ Current date and time: ${localStr} (${tz}${offsetLabel ? `, ${offsetLabel}` : ""
         };
       }
       /**
-       * Drop lower-priority converse sections until system ≤ budgetTokens.
-       * Uses the real token counter (not chars/4) so packing matches afford checks.
-       * Order: team norms/announcements → Search Strategy → roster → other dynamics.
+       * Tier-3 tail entries that MUST survive converse budget trim.
+       * Date/locale are small (~50–150 tok) but critical for time-aware replies;
+       * last-resort `dynamic.pop()` previously deleted them first because they
+       * sit at the end of the dynamic array (PROMPT-ENGINEERING §2 Tier 3).
+       */
+      isProtectedDynamicTail(entry) {
+        return entry.includes("Current date and time:") || entry.startsWith("User locale:") || entry.startsWith("User language:");
+      }
+      /** Keep Interaction Mode so the agent knows which scenario it is in. */
+      isProtectedSemiStableTail(entry) {
+        return entry.includes("## Current Interaction Mode");
+      }
+      /**
+       * Failsafe only (Afford.S3): drop lower-priority converse sections until
+       * system ≤ budgetTokens. Normal packing MUST stay under budget via profile
+       * gates and injection caps — if this method runs, log a warning so upstream
+       * bloat can be fixed. Never drops Current date/time, user locale/language,
+       * or Interaction Mode.
        */
       trimConverseSystemBudget(stable, semiStable, dynamic, budgetTokens) {
         const totalTokens = () => estimateTokens([stable.join("\n"), semiStable.join("\n"), dynamic.join("\n")].join("\n"), this.tokenCounter);
-        if (totalTokens() <= budgetTokens)
+        const before2 = totalTokens();
+        if (before2 <= budgetTokens)
           return;
+        log8.warn("Converse system over budget after assemble \u2014 failsafe trim engaged", {
+          systemTokens: before2,
+          budgetTokens,
+          hint: "Fix upstream injection caps / profile gates; trim must stay rare"
+        });
         const dropHeadingBlock = (arr, heading) => {
           const start = arr.findIndex((s2) => s2.includes(heading));
           if (start < 0)
@@ -8865,64 +8934,87 @@ Current date and time: ${localStr} (${tz}${offsetLabel ? `, ${offsetLabel}` : ""
           arr.splice(start, end - start);
           return true;
         };
-        dropHeadingBlock(semiStable, "## Team Announcements");
-        if (totalTokens() <= budgetTokens)
-          return;
-        dropHeadingBlock(semiStable, "## Team Working Norms");
-        if (totalTokens() <= budgetTokens)
-          return;
-        dropHeadingBlock(stable, "## Search & Exploration Strategy");
-        if (totalTokens() <= budgetTokens)
-          return;
-        dropHeadingBlock(semiStable, "### Colleagues");
-        dropHeadingBlock(semiStable, "### Your Team");
-        dropHeadingBlock(dynamic, "### Colleagues");
-        dropHeadingBlock(dynamic, "## Colleague Status");
-        if (totalTokens() <= budgetTokens)
-          return;
-        const dynamicDropOrder = [
-          "## Channel Context",
-          "## Active Workflows",
-          "## Mailbox",
-          "## Current Project",
-          "## Shared Deliverables",
-          "## Task Board"
-        ];
-        for (const h of dynamicDropOrder) {
-          if (totalTokens() <= budgetTokens)
-            return;
-          dropHeadingBlock(dynamic, h);
-        }
-        while (totalTokens() > budgetTokens && dynamic.length > 0) {
-          dynamic.pop();
-        }
-        while (totalTokens() > budgetTokens && semiStable.length > 1) {
-          semiStable.pop();
+        const protectedDynamic = [];
+        while (dynamic.length > 0 && this.isProtectedDynamicTail(dynamic[dynamic.length - 1])) {
+          protectedDynamic.unshift(dynamic.pop());
         }
-        if (totalTokens() > budgetTokens) {
-          const joined = stable.join("\n");
-          let lo = 0;
-          let hi = joined.length;
-          let best = "";
-          while (lo <= hi) {
-            const mid = Math.floor((lo + hi) / 2);
-            const candidate = `${joined.slice(0, mid)}
+        const protectedTok = protectedDynamic.length ? estimateTokens(protectedDynamic.join("\n"), this.tokenCounter) : 0;
+        const workingBudget = Math.max(1e3, budgetTokens - protectedTok);
+        try {
+          dropHeadingBlock(semiStable, "## Team Announcements");
+          if (totalTokens() <= workingBudget)
+            return;
+          dropHeadingBlock(semiStable, "## Team Working Norms");
+          if (totalTokens() <= workingBudget)
+            return;
+          dropHeadingBlock(stable, "## Search & Exploration Strategy");
+          if (totalTokens() <= workingBudget)
+            return;
+          dropHeadingBlock(semiStable, "### Colleagues");
+          dropHeadingBlock(semiStable, "### Your Team");
+          dropHeadingBlock(dynamic, "### Colleagues");
+          dropHeadingBlock(dynamic, "## Colleague Status");
+          if (totalTokens() <= workingBudget)
+            return;
+          const dynamicDropOrder = [
+            "## Channel Context",
+            "## Channel History",
+            "## Available Workflows",
+            "## Active Workflows",
+            "## Mailbox",
+            "## Current Project",
+            "## Shared Deliverables",
+            "## Task Board",
+            "## Team Status"
+          ];
+          for (const h of dynamicDropOrder) {
+            if (totalTokens() <= workingBudget)
+              return;
+            dropHeadingBlock(dynamic, h);
+          }
+          while (totalTokens() > workingBudget && dynamic.length > 0) {
+            dynamic.pop();
+          }
+          while (totalTokens() > workingBudget && semiStable.length > 1) {
+            const last = semiStable[semiStable.length - 1];
+            if (this.isProtectedSemiStableTail(last)) {
+              let idx = semiStable.length - 2;
+              while (idx >= 1 && this.isProtectedSemiStableTail(semiStable[idx]))
+                idx--;
+              if (idx < 1)
+                break;
+              semiStable.splice(idx, 1);
+              continue;
+            }
+            semiStable.pop();
+          }
+          if (totalTokens() > workingBudget) {
+            const joined = stable.join("\n");
+            let lo = 0;
+            let hi = joined.length;
+            let best = "";
+            while (lo <= hi) {
+              const mid = Math.floor((lo + hi) / 2);
+              const candidate = `${joined.slice(0, mid)}
 
 _[system trimmed to ${budgetTokens} tok converse budget]_`;
-            const other = [semiStable.join("\n"), dynamic.join("\n")].join("\n");
-            const tok = estimateTokens(`${candidate}
+              const other = [semiStable.join("\n"), dynamic.join("\n")].join("\n");
+              const tok = estimateTokens(`${candidate}
 ${other}`, this.tokenCounter);
-            if (tok <= budgetTokens) {
-              best = candidate;
-              lo = mid + 1;
-            } else {
-              hi = mid - 1;
+              if (tok <= workingBudget) {
+                best = candidate;
+                lo = mid + 1;
+              } else {
+                hi = mid - 1;
+              }
             }
-          }
-          stable.length = 0;
-          stable.push(best || `${joined.slice(0, Math.max(0, budgetTokens * 2))}
+            stable.length = 0;
+            stable.push(best || `${joined.slice(0, Math.max(0, workingBudget * 2))}
 
 _[system trimmed]_`);
+          }
+        } finally {
+          dynamic.push(...protectedDynamic);
         }
       }
       /** Human-readable language name for a BCP-47 locale (e.g. 'zh-CN' → 'Chinese (China)'). */
@@ -9034,11 +9126,12 @@ ${ctx.mergedContent.slice(0, SYSTEM_MAILBOX_MERGED_CHARS)}`);
         return lines.join("\n");
       }
       /**
-       * L3 scenario-triggered policy blocks — long checklists that chat/heartbeat
-       * must not carry. Loaded for task execution, review, and related modes.
+       * L3 scenario-triggered policy blocks — long checklists that converse/reflex
+       * must not carry. Caller gates on execute/govern profile; this further
+       * restricts to task/review/deliberation scenarios.
        */
       buildScenarioPolicyBlocks(scenario) {
-        const needsExecutionPolicies = scenario === "task_execution" || scenario === "review" || scenario === "deliberation" || scenario === "comment_response";
+        const needsExecutionPolicies = scenario === "task_execution" || scenario === "review" || scenario === "deliberation";
         if (!needsExecutionPolicies)
           return void 0;
         const lines = [];
@@ -69365,7 +69458,7 @@ var init_openai = __esm({
         this.baseUrl = config?.baseUrl ?? "https://api.openai.com";
         this.maxTokens = config?.maxTokens ?? DEFAULT_REQUEST_MAX_TOKENS;
         this.chatTimeoutMs = config?.timeoutMs ?? 9e4;
-        this.streamTimeoutMs = config?.timeoutMs ?? 18e4;
+        this.streamTimeoutMs = config?.streamTimeoutMs ?? 18e4;
         this.tokenResolver = tokenResolver;
       }
       configure(config) {
@@ -69377,10 +69470,10 @@ var init_openai = __esm({
           this.baseUrl = config.baseUrl;
         if (config.maxTokens)
           this.maxTokens = config.maxTokens;
-        if (config.timeoutMs) {
+        if (config.timeoutMs)
           this.chatTimeoutMs = config.timeoutMs;
-          this.streamTimeoutMs = config.timeoutMs;
-        }
+        if (config.streamTimeoutMs)
+          this.streamTimeoutMs = config.streamTimeoutMs;
       }
       setTokenResolver(resolver) {
         this.tokenResolver = resolver;
@@ -69525,11 +69618,22 @@ var init_openai = __esm({
         const endpoint = /\/v\d+$/.test(base) ? `${base}/chat/completions` : `${base}/v1/chat/completions`;
         const controller = new AbortController();
         const STREAM_HARD_TIMEOUT_MS2 = 15 * 6e4;
-        let idleTimeout = setTimeout(() => controller.abort(), this.streamTimeoutMs);
-        const hardTimeout = setTimeout(() => controller.abort(), STREAM_HARD_TIMEOUT_MS2);
+        let idleTimedOut = false;
+        let hardTimedOut = false;
+        let idleTimeout = setTimeout(() => {
+          idleTimedOut = true;
+          controller.abort();
+        }, this.streamTimeoutMs);
+        const hardTimeout = setTimeout(() => {
+          hardTimedOut = true;
+          controller.abort();
+        }, STREAM_HARD_TIMEOUT_MS2);
         const bumpIdleTimeout = () => {
           clearTimeout(idleTimeout);
-          idleTimeout = setTimeout(() => controller.abort(), this.streamTimeoutMs);
+          idleTimeout = setTimeout(() => {
+            idleTimedOut = true;
+            controller.abort();
+          }, this.streamTimeoutMs);
         };
         const clearStreamTimeouts = () => {
           clearTimeout(idleTimeout);
@@ -69640,6 +69744,18 @@ var init_openai = __esm({
               }
             }
           }
+        } catch (err) {
+          clearStreamTimeouts();
+          const hasPartial = content.length > 0 || reasoningContent.length > 0 || toolCalls.size > 0;
+          if (idleTimedOut && hasPartial) {
+            toolCalls.clear();
+            finishReason = "max_tokens";
+          } else if (idleTimedOut || hardTimedOut) {
+            const kind = hardTimedOut ? "hard" : "idle";
+            throw new Error(`OpenAI stream ${kind} timeout after ${hardTimedOut ? STREAM_HARD_TIMEOUT_MS2 : this.streamTimeoutMs}ms`);
+          } else {
+            throw err;
+          }
         } finally {
           clearStreamTimeouts();
         }
@@ -71515,6 +71631,26 @@ function stripMarkusNamespace(modelId) {
   const id = (modelId ?? "").trim();
   return id.replace(/^markus\//i, "");
 }
+function parseRetryAfterMs(res) {
+  const raw = res.headers.get("retry-after");
+  if (!raw)
+    return null;
+  const asSec = Number(raw);
+  if (Number.isFinite(asSec) && asSec >= 0) {
+    return Math.min(asSec * 1e3, MAX_RETRY_AFTER_MS);
+  }
+  const asDate = Date.parse(raw);
+  if (Number.isFinite(asDate)) {
+    return Math.min(Math.max(0, asDate - Date.now()), MAX_RETRY_AFTER_MS);
+  }
+  return null;
+}
+function retryDelayMs(res, attempt) {
+  const fromHeader = res ? parseRetryAfterMs(res) : null;
+  if (fromHeader !== null)
+    return fromHeader;
+  return RETRY_BASE_DELAY_MS * Math.pow(2, attempt);
+}
 function fireCreditExhaustedEvent() {
   const g = globalThis;
   if (typeof g.dispatchEvent !== "function")
@@ -71598,7 +71734,7 @@ function detectAudioFormat(audio) {
   }
   return "wav";
 }
-var log52, CUCache, DEFAULT_OR_BASE_URL, DEFAULT_MODEL, CLIENT_ID, CHAT_TIMEOUT_MS, STREAM_TIMEOUT_MS, STREAM_HARD_TIMEOUT_MS, MAX_RETRIES, RETRY_BASE_DELAY_MS, CREDIT_MUTE_KEY, _lastCreditNotifTs, cachedModelList, modelListExpiry, MODEL_LIST_TTL_MS, MarkusProvider, FALLBACK_MODELS;
+var log52, CUCache, DEFAULT_OR_BASE_URL, DEFAULT_MODEL, CLIENT_ID, CHAT_TIMEOUT_MS, STREAM_TIMEOUT_MS, STREAM_HARD_TIMEOUT_MS, MAX_RETRIES, RETRY_BASE_DELAY_MS, MAX_RETRY_AFTER_MS, CREDIT_MUTE_KEY, _lastCreditNotifTs, cachedModelList, modelListExpiry, MODEL_LIST_TTL_MS, MarkusProvider, FALLBACK_MODELS;
 var init_markus_provider = __esm({
   "../core/dist/llm/markus-provider.js"() {
     "use strict";
@@ -71637,6 +71773,7 @@ var init_markus_provider = __esm({
     STREAM_HARD_TIMEOUT_MS = 15 * 6e4;
     MAX_RETRIES = 3;
     RETRY_BASE_DELAY_MS = 500;
+    MAX_RETRY_AFTER_MS = 6e4;
     CREDIT_MUTE_KEY = "markus:credit-notif-muted";
     _lastCreditNotifTs = 0;
     cachedModelList = null;
@@ -71667,7 +71804,7 @@ var init_markus_provider = __esm({
         this.baseUrl = config?.baseUrl ?? DEFAULT_OR_BASE_URL;
         this.maxTokens = config?.maxTokens;
         this.chatTimeoutMs = config?.timeoutMs ?? CHAT_TIMEOUT_MS;
-        this.streamTimeoutMs = config?.timeoutMs ?? STREAM_TIMEOUT_MS;
+        this.streamTimeoutMs = config?.streamTimeoutMs ?? STREAM_TIMEOUT_MS;
         this.modelsUrl = config?.modelsUrl ?? process.env["MARKUS_MODELS_URL"] ?? "";
         this.hubUrl = config?.hubUrl ?? process.env["MARKUS_HUB_URL"] ?? "";
         this.hubToken = config?.hubToken ?? process.env["MARKUS_HUB_TOKEN"] ?? "";
@@ -71691,10 +71828,10 @@ var init_markus_provider = __esm({
           this.hubUrl = config.hubUrl;
         if (config.hubToken !== void 0)
           this.hubToken = config.hubToken;
-        if (config.timeoutMs) {
+        if (config.timeoutMs)
           this.chatTimeoutMs = config.timeoutMs;
-          this.streamTimeoutMs = config.timeoutMs;
-        }
+        if (config.streamTimeoutMs)
+          this.streamTimeoutMs = config.streamTimeoutMs;
       }
       /** Whether OpenRouter credentials are available. */
       hasOpenRouterCreds() {
@@ -72160,6 +72297,7 @@ var init_markus_provider = __esm({
         }
         const decoder = new TextDecoder();
         let buffer = "";
+        let midStreamRateLimited = false;
         try {
           while (true) {
             const { done, value } = await reader.read();
@@ -72169,6 +72307,7 @@ var init_markus_provider = __esm({
             buffer += decoder.decode(value, { stream: true });
             const lines = buffer.split("\n");
             buffer = lines.pop() ?? "";
+            let stopStream = false;
             for (const line of lines) {
               const trimmed = line.trim();
               if (!trimmed || trimmed === "data: [DONE]")
@@ -72179,6 +72318,19 @@ var init_markus_provider = __esm({
                 const chunk = JSON.parse(trimmed.slice(6));
                 const choice = chunk.choices?.[0];
                 const delta = choice?.delta;
+                const streamErr = chunk.error ?? null;
+                const finishRaw = choice?.finish_reason !== null ? String(choice?.finish_reason ?? "") : "";
+                if (streamErr || finishRaw === "error") {
+                  const code = Number(streamErr?.code ?? 0);
+                  const errMsg = String(streamErr?.message ?? "Rate limit exceeded");
+                  const isRateLimit = code === 429 || /rate.?limit/i.test(errMsg);
+                  if (isRateLimit) {
+                    midStreamRateLimited = true;
+                    stopStream = true;
+                    break;
+                  }
+                  throw new Error(`Markus stream error ${code || "unknown"}: ${errMsg}`);
+                }
                 const deltaReasoning = extractDeltaReasoning(delta);
                 if (deltaReasoning) {
                   reasoningContent += deltaReasoning;
@@ -72225,18 +72377,47 @@ var init_markus_provider = __esm({
                     this.recordCostUsd(chunk.usage);
                   }
                 }
-              } catch {
+              } catch (parseOrStreamErr) {
+                if (parseOrStreamErr instanceof Error && parseOrStreamErr.message.startsWith("Markus stream error")) {
+                  throw parseOrStreamErr;
+                }
               }
             }
+            if (stopStream)
+              break;
+          }
+          if (midStreamRateLimited) {
+            const hasPartial = content.length > 0 || reasoningContent.length > 0 || toolCallsAcc.size > 0;
+            if (hasPartial) {
+              log52.warn("OpenRouter mid-stream rate limit \u2014 returning partial for auto-continuation", {
+                contentChars: content.length
+              });
+              toolCallsAcc.clear();
+              finishReason = "max_tokens";
+            } else {
+              throw new Error("MARKUS_RATE_LIMITED: Rate limit exceeded (mid-stream)");
+            }
           }
         } catch (err) {
           clearStreamTimeouts();
           const msg = err instanceof Error ? err.message : String(err);
-          if (idleTimedOut || hardTimedOut || /aborted/i.test(msg)) {
+          const hasPartial = content.length > 0 || reasoningContent.length > 0 || toolCallsAcc.size > 0;
+          if (idleTimedOut && hasPartial) {
+            log52.warn("Markus stream idle timeout with partial content \u2014 returning for auto-continuation", {
+              idleMs: this.streamTimeoutMs,
+              contentChars: content.length,
+              droppedToolCalls: toolCallsAcc.size
+            });
+            toolCallsAcc.clear();
+            finishReason = "max_tokens";
+          } else if (idleTimedOut || hardTimedOut) {
             const kind = hardTimedOut ? "hard" : "idle";
             throw new Error(`Markus stream ${kind} timeout after ${hardTimedOut ? STREAM_HARD_TIMEOUT_MS : this.streamTimeoutMs}ms`);
+          } else if (signal?.aborted || /aborted/i.test(msg)) {
+            throw err instanceof Error ? err : new Error(msg);
+          } else {
+            throw err;
           }
-          throw err;
         } finally {
           clearStreamTimeouts();
         }
@@ -72347,32 +72528,48 @@ var init_markus_provider = __esm({
       }
       /**
        * Fetch with exponential backoff retry.
-       * Does NOT retry on 4xx errors (billing/auth are not transient).
-       * Only retries on 5xx / network issues.
+       *
+       * OpenRouter guidance (https://openrouter.ai/docs/api_reference/limits):
+       * - 429 / 503 are transient — retry with exponential backoff; honor `Retry-After`
+       * - Other 4xx (auth/billing) are not retried
+       * - 5xx / network errors are retried (unless skipRetry — used after stream body
+       *   would already be in flight; 429/503 still retry because they fail before body)
        */
       async fetchWithRetry(url, init, skipRetry = false, retries = MAX_RETRIES) {
         let lastError;
         for (let attempt = 0; attempt < retries; attempt++) {
+          let res;
           try {
-            const res = await fetch(url, init);
-            if (res.status >= 400 && res.status < 500) {
-              return res;
-            }
+            res = await fetch(url, init);
             if (res.ok)
               return res;
-            log52.warn(`Markus proxy error ${res.status} (attempt ${attempt + 1}/${retries})`);
+            const retriableStatus = res.status === 429 || res.status === 503;
+            if (res.status >= 400 && res.status < 500 && !retriableStatus) {
+              return res;
+            }
+            log52.warn(`OpenRouter/Markus HTTP ${res.status} (attempt ${attempt + 1}/${retries})`, {
+              retryAfter: res.headers.get("retry-after"),
+              rateLimitRemaining: res.headers.get("x-ratelimit-remaining"),
+              rateLimitReset: res.headers.get("x-ratelimit-reset")
+            });
             lastError = new Error(`Markus proxy error ${res.status}`);
-            if (skipRetry)
+            if (skipRetry && !retriableStatus)
+              return res;
+            if (attempt >= retries - 1)
               return res;
             const errText = await res.text().catch(() => "");
             if (errText)
               log52.warn("Response body", { body: errText.slice(0, 200) });
+            const delay = retryDelayMs(res, attempt);
+            log52.info(`Waiting ${delay}ms before retry (Retry-After / backoff)`);
+            await new Promise((r) => setTimeout(r, delay));
+            continue;
           } catch (err) {
             lastError = err instanceof Error ? err : new Error(String(err));
             log52.warn(`Markus proxy network error (attempt ${attempt + 1}/${retries})`, { error: lastError.message });
-          }
-          if (attempt < retries - 1) {
-            const delay = RETRY_BASE_DELAY_MS * Math.pow(2, attempt);
+            if (skipRetry || attempt >= retries - 1)
+              break;
+            const delay = retryDelayMs(void 0, attempt);
             await new Promise((r) => setTimeout(r, delay));
           }
         }
@@ -235912,6 +236109,7 @@ async function createServices(config) {
   const providerConfigs = {};
   let defaultProvider = config.llm.defaultProvider;
   const llmTimeoutMs = Number(process.env["LLM_TIMEOUT_MS"]) || config.llm.timeoutMs || void 0;
+  const llmStreamTimeoutMs = Number(process.env["LLM_STREAM_TIMEOUT_MS"]) || void 0;
   const providerMeta = new Map(PROVIDERS.map((p) => [p.id, p]));
   const envPrefix = (id) => id.toUpperCase().replace(/-/g, "_");
   const markusCfg = config.llm.providers["markus"] ?? {};
@@ -235982,7 +236180,8 @@ async function createServices(config) {
       ...markusModelsUrl ? { modelsUrl: markusModelsUrl } : {},
       hubUrl,
       ...hubToken ? { hubToken } : {},
-      timeoutMs: llmTimeoutMs
+      timeoutMs: llmTimeoutMs,
+      ...llmStreamTimeoutMs ? { streamTimeoutMs: llmStreamTimeoutMs } : {}
     };
   } else if (config.llm.defaultProvider === "markus") {
     log103.warn(
@@ -236006,7 +236205,8 @@ async function createServices(config) {
           provider: "ollama",
           model: model2,
           ...baseUrl2 ? { baseUrl: baseUrl2 } : {},
-          timeoutMs: llmTimeoutMs
+          timeoutMs: llmTimeoutMs,
+          ...llmStreamTimeoutMs ? { streamTimeoutMs: llmStreamTimeoutMs } : {}
         };
       }
       continue;
@@ -236021,7 +236221,8 @@ async function createServices(config) {
       model,
       apiKey,
       ...baseUrl ? { baseUrl } : {},
-      timeoutMs: llmTimeoutMs
+      timeoutMs: llmTimeoutMs,
+      ...llmStreamTimeoutMs ? { streamTimeoutMs: llmStreamTimeoutMs } : {}
     };
   }
   if (!providerConfigs[defaultProvider]) {