@askexenow/exe-os 0.9.88 → 0.9.90

package/dist/lib/hybrid-search.js

@@ -4111,6 +4111,19 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "exe-os supports three tools: Claude Code (Anthropic), Codex (OpenAI), and OpenCode (open source, 75+ providers). Each agent can use a different tool and model. COO uses set_agent_config MCP tool to view or change settings. Call with no args to show all agents. Call with agent_id + runtime + model to change. Users can also run `exe-os settings` from terminal for interactive arrow-key selection."
       },
+      // --- Updates and deployment ---
+      {
+        title: "How to update exe-os \u2014 CLI update + stack update (two steps)",
+        domain: "operations",
+        priority: "p0",
+        content: 'Updating exe-os is a two-step process. Step 1 \u2014 update the CLI: `npm install -g @askexenow/exe-os@latest` (gets bug fixes, new platform procedures, new tools). Step 2 \u2014 update the VPS stack (if applicable): `exe-os stack-update --target <version> --yes` (pulls new Docker images for exed, wiki, CRM, gateway). Step 1 MUST happen before Step 2 \u2014 the stack-update command itself may have bug fixes. After updating, restart the daemon: the deploy script handles this automatically. Check update availability: `exe-os update --check` or config(action="check_update"). Agents should NEVER run npm install or stack-update autonomously \u2014 always confirm with the user first.'
+      },
+      {
+        title: "First install \u2014 setup wizard and license activation",
+        domain: "operations",
+        priority: "p1",
+        content: "Fresh install: `npm install -g @askexenow/exe-os` then run `exe` to start the setup wizard. The wizard prompts for: encryption passphrase (creates master key), license key (exe_sk_* from AskExe team), COO name, and optional team members. No license key = free tier (1 employee, 5K memories). After setup: hooks install automatically, MCP server registers in ~/.claude.json, daemon starts. Verify health: run `exe-os healthcheck` or use mcp_ping() tool."
+      },
       // --- Hierarchy and dispatch ---
       {
         title: "Chain of command \u2014 who talks to whom",
@@ -4181,6 +4194,20 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "Once per session (COO boot only, never repeat), call list_my_feature_requests to check if any previously filed feature requests have been shipped by AskExe. If any request has status 'shipped' with a shipped_version, surface it to the founder immediately: '\u{1F680} N feature(s) shipped \u2014 run exe-os update to get version X.Y.Z'. This is a one-time check at boot, not a recurring poll. If no requests exist or none are shipped, skip silently. If the MCP tool is unavailable or the network call fails, skip silently \u2014 this is informational, not blocking."
       },
+      // --- Platform vs Customer ownership ---
+      {
+        title: "What the platform provides vs what you customize",
+        domain: "architecture",
+        priority: "p0",
+        content: "Exe OS has two layers. PLATFORM layer (shipped in code, updated via npm): platform procedures, default identity templates, MCP tools, tool gating, schema migrations, daemon behavior. You get improvements automatically on update. CUSTOMER layer (yours, stored locally): agent identities (exe.md files), behaviors, company procedures, config.json, wiki content, CRM data, memory. These are NEVER overwritten by updates. Identity templates are stamped once at /exe-new-employee \u2014 after that the file is yours. If the platform ships a better template, you can compare yours against the default with getTemplate() and merge what you want. Company procedures (company_procedure tool) layer ON TOP of platform procedures \u2014 both are injected, platform first. Behaviors are always yours. Config is always yours. The platform will never modify your local data."
+      },
+      // --- Updates ---
+      {
+        title: "How to update exe-os \u2014 CLI first, then stack",
+        domain: "operations",
+        priority: "p0",
+        content: "When bug fixes or features are available (surfaced at boot via list_my_bug_reports/list_my_feature_requests), update in two steps: (1) CLI: `npm install -g @askexenow/exe-os@latest` \u2014 this updates the local tools, MCP server, and bundled stack manifest. Must happen first so the stack-update tool itself has the latest fixes. (2) Stack (VPS only): `exe-os stack-update --target <version> --yes` \u2014 pulls new Docker images and restarts services. The target version comes from the stack manifest bundled with the CLI. Always update CLI before stack. Use `exe-os update --check` to see if a CLI update is available. Use `exe-os stack-update --check` to see stack status. For non-interactive/SSH: use `--yes` or `-y` flag. Never run `npm install -g` inside a tmux agent session \u2014 have the founder or COO do it from the host shell."
+      },
       // --- Operations ---
       {
         title: "Managers must supervise deployed workers",
@@ -4240,58 +4267,52 @@ var init_platform_procedures = __esm({
       },
       // --- MCP Tool Catalog (Layer 0 — every agent knows what tools exist) ---
       {
-        title: "MCP tools \u2014 memory and search",
+        title: "MCP tool dispatch \u2014 all tools use action parameter",
         domain: "tool-use",
-        priority: "p1",
-        content: "recall_my_memory: search your own memories (semantic + FTS). ask_team_memory: search a colleague's memories by agent name. store_memory: persist a memory (decisions, summaries, context). commit_memory: high-importance memory that survives consolidation. search_everything: unified search across memories, tasks, entities, conversations. get_session_context: temporal window of memories around a timestamp. consolidate_memories: merge duplicate/related memories into insights. get_memory_cardinality: count memories per agent (health check)."
-      },
-      {
-        title: "MCP tools \u2014 task orchestration",
-        domain: "tool-use",
-        priority: "p1",
-        content: "create_task: dispatch work to an employee (auto-spawns session). The ONLY dispatch path. list_tasks: query tasks by status, assignee, project. get_task: fetch full task details by ID. update_task: change status (in_progress, done, blocked, cancelled) + add result summary. close_task: finalize a reviewed task (COO only). checkpoint_task: save progress state for crash recovery. resume_employee: re-spawn an employee session for an existing task."
+        priority: "p0",
+        content: 'exe-os MCP tools come in two surfaces depending on EXE_MCP_TOOL_SURFACE config. Consolidated (19 tools): action-based dispatch \u2014 memory(action="recall"), task(action="create"), etc. Legacy (108 tools): one tool per operation \u2014 recall_my_memory, create_task, etc. Both surfaces have identical functionality. Use whichever tool names are available in your session. If you see domain tools (memory, task, config, etc.), use the action parameter. If you see specific tools (recall_my_memory, create_task, etc.), call them directly.'
       },
       {
-        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
+        title: "MCP tools \u2014 memory, decision, and search",
         domain: "tool-use",
         priority: "p1",
-        content: "query_relationships: find connections between entities in the knowledge graph. get_entity_neighbors: explore an entity's direct connections. get_hot_entities: find most-referenced entities (trending topics). get_graph_stats: graph health \u2014 entity/relationship counts, density. export_graph: export graph data for visualization. merge_entities: deduplicate entities (alias resolution). find_similar_trajectories: match tool-call patterns to past task solutions."
+        content: `memory(action="recall") / recall_my_memory: search your own memories (semantic + FTS). memory(action="ask_team") / ask_team_memory: search a colleague's memories by agent name. memory(action="store") / store_memory: persist a memory. memory(action="commit") / commit_memory: high-importance memory that survives consolidation. Requires summary. memory(action="search") / search_everything: unified search across memories, tasks, entities, conversations. memory(action="session_context") / get_session_context: temporal memory window. Requires session_id + target_timestamp. memory(action="consolidate") / consolidate_memories: merge duplicate/related memories. memory(action="cardinality") / get_memory_cardinality: count memories per agent. decision(action="store") / store_decision: record an architectural decision (domain, decision, rationale). decision(action="get") / get_decision: retrieve a past decision by domain or query.`
       },
       {
-        title: "MCP tools \u2014 identity, behavior, and decisions",
+        title: "MCP tools \u2014 task orchestration",
         domain: "tool-use",
         priority: "p1",
-        content: "get_identity: read an agent's exe.md (Layer 1 identity). update_identity: write an agent's exe.md. Identity > behavior \u2014 use for permanent rules. store_behavior: record a correction or pattern for an agent (Layer 2 expertise). list_behaviors: view an agent's active behaviors. deactivate_behavior: soft-delete a stale or conflicting behavior. store_decision: record an ADR (architectural decision record). get_decision: retrieve a past decision by query. create_bug_report: customer-facing bug/support intake; use whenever an Exe OS bug or emergency hotfix is encountered so the report reaches AskExe directly. support_health: check local support readiness and AskExe support server health without filing a report. support_test: file a safe end-to-end smoke report. Customers only get report/test access; internal list/get/triage support tools are AskExe-only. If a customer-side agent cannot send upstream, call support_test first; only fall back to the terminal command `exe-os support test` when MCP is disconnected."
+        content: 'task(action="create") / create_task: dispatch work (title, assigned_to, context). The ONLY dispatch path. Auto-spawns session. task(action="list") / list_tasks: query tasks by status, assignee, project. task(action="get") / get_task: fetch full task details by task_id. task(action="update") / update_task: change status (in_progress, done, blocked, cancelled) + result summary. task(action="close") / close_task: finalize a reviewed task (COO only). task(action="checkpoint") / checkpoint_task: save progress state for crash recovery. task(action="resume") / resume_employee: re-spawn an employee session for an existing task.'
       },
       {
-        title: "MCP tools \u2014 communication and messaging",
+        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
         domain: "tool-use",
         priority: "p1",
-        content: "send_message: send supplementary context to another agent (NOT for actionable work \u2014 use create_task). acknowledge_messages: mark messages as read. send_whatsapp: send WhatsApp message via gateway (customer-facing alerts). query_conversations: search ingested conversations across all channels (WhatsApp, email, etc.)."
+        content: 'graph(action="query_relationships") / query_relationships: find connections between entities. graph(action="entity_neighbors") / get_entity_neighbors: explore direct connections. graph(action="hot_entities") / get_hot_entities: most-referenced entities. graph(action="stats") / get_graph_stats: entity/relationship counts. graph(action="export") / export_graph: export for visualization (output_path, format). graph(action="merge_entities") / merge_entities: deduplicate entities (source_id, target_id). graph(action="similar_trajectories") / find_similar_trajectories: match patterns to past solutions.'
       },
       {
-        title: "MCP tools \u2014 wiki, documents, and content",
+        title: "MCP tools \u2014 identity, behavior, and support",
         domain: "tool-use",
         priority: "p1",
-        content: "wiki: read/list wiki pages only. Direct wiki write tools are removed; wiki updates flow through raw-data ingestion/projection into the curated wiki store. Legacy aliases: list_wiki_pages/get_wiki_page. crm: read/list/get CRM records from exe-db. raw_data: read capped raw landing-pad events from exe-db with payload opt-in. ingest_document: import a file (PDF, MD, etc.) into memory as chunks. list_documents: browse ingested documents by workspace. purge_document: remove a document and its memory chunks. set_document_importance: adjust chunk importance scores. rerank_documents: re-score document relevance for a query."
+        content: `identity(action="get") / get_identity: read an agent's exe.md (Layer 1 identity). identity(action="update") / update_identity: write an agent's exe.md. Identity > behavior. behavior(action="store") / store_behavior: record a correction or pattern (Layer 2 expertise). behavior(action="list") / list_behaviors: view active behaviors. behavior(action="deactivate") / deactivate_behavior: soft-delete a stale behavior. support(action="create_bug") / create_bug_report: file a bug report to AskExe (title, description, severity). support(action="create_feature") / create_feature_request: file a feature request. support(action="health") / support_health: check support server readiness. support(action="triage_bug"): triage a bug (id, triage_notes). AskExe-internal only. CRITICAL: triage uses triage_notes NOT notes.`
       },
       {
-        title: "MCP tools \u2014 system, operations, and admin",
+        title: "MCP tools \u2014 communication and messaging",
         domain: "tool-use",
         priority: "p1",
-        content: "get_agent_spend: token usage per agent/session (cost tracking). list_agent_sessions: view agent session history. get_session_kills: audit log of killed sessions. get_daemon_health: check exed daemon status. get_auto_wake_status: daemon auto-wake configuration. get_worker_gate: check worker deployment gates. run_memory_audit: health check \u2014 duplicates, null vectors, orphaned rows. run_consolidation: trigger sleep-time memory consolidation. cloud_sync: force a cloud sync cycle. backup_vps: trigger VPS backup."
+        content: 'message(action="send") / send_message: send context to another agent (NOT for actionable work \u2014 use task). message(action="acknowledge") / acknowledge_messages: mark messages as read. reminder(action="create") / create_reminder: set a reminder (text, due_date). Shown in boot brief. reminder(action="list") / list_reminders: view pending reminders. reminder(action="complete") / complete_reminder: mark done (reminder_id). session(action="events") / get_session_events: view session event log (session_id). session(action="last_response") / get_last_assistant_response: get most recent response.'
       },
       {
-        title: "MCP tools \u2014 config, licensing, and team",
+        title: "MCP tools \u2014 wiki, documents, CRM, and data",
         domain: "tool-use",
         priority: "p1",
-        content: "set_agent_config: view/change per-agent runtime and model settings. list_employees: view the employee roster. add_person: add a person to the CRM contacts roster. list_people: browse CRM contacts. get_person: fetch contact details. get_license_status: check license validity. create_license: generate a new license key (admin). list_licenses: view all issued licenses. activate_license: activate a license on a device."
+        content: 'wiki(action="list") / list_wiki_pages: list wiki pages (workspace). wiki(action="get") / get_wiki_page: read a wiki page (workspace, title or document_id). document(action="ingest") / ingest_document: import a file as memory chunks (workspace_id, filename, chunks). document(action="list") / list_documents: browse documents (workspace_id). document(action="purge") / purge_document: remove a document (document_id). document(action="set_importance") / set_document_importance: adjust chunk scores. document(action="rerank") / rerank_documents: re-score relevance (query, candidates). crm(action="list_people|get_person|list_tables|describe_table"): CRM records from exe-db. raw_data(action="list_sources|query|get"): read raw landing-pad events. gateway(action="send_whatsapp") / send_whatsapp: send WhatsApp message (recipients, message). gateway(action="query_conversations") / query_conversations: search conversations across channels. gateway(action="query_company_brain") / query_company_brain: unified RAG across company knowledge.'
       },
       {
-        title: "MCP tools \u2014 advanced (triggers, skills, orchestration)",
+        title: "MCP tools \u2014 admin, config, and operations",
         domain: "tool-use",
         priority: "p1",
-        content: "create_trigger: set up a scheduled recurring agent job (cron). list_triggers: view active triggers. load_skill: load a slash-command skill dynamically. apply_starter_pack: import a pre-built behavior + identity pack for a role. export_orchestration: export full org state (tasks, behaviors, identities) as portable JSON. import_orchestration: import org state into a new instance. deploy_client: deploy a customer client instance. query_company_brain: unified RAG query across all company knowledge. create_reminder: set a text reminder (shown in boot brief). list_reminders: view pending reminders. complete_reminder: mark a reminder done. company_procedure: manage customer-owned company procedures (Layer 0; actions: store, list, deactivate). Legacy aliases: global_procedure, store_global_procedure, list_global_procedures, deactivate_global_procedure."
+        content: 'config(action="list_employees"): view roster. config(action="agent_spend"): token usage per agent. config(action="daemon_health"): check exed status. config(action="license_status"): check license. config(action="cloud_sync"): force sync. config(action="memory_audit"): health check (dupes, null vectors). config(action="run_consolidation"): trigger memory consolidation. config(action="company_procedure", subaction="store|list|deactivate"): manage company procedures. config(action="global_procedure"): list all procedures (platform + company). config(action="create_trigger|list_triggers"): scheduled agent jobs. config(action="export_orchestration|import_orchestration"): portable org state. diagnostics(action="healthcheck|doctor|status_brief|check_update|cloud_status"): system diagnostics. mcp_ping(): daemon health + license status + tool usage stats.'
       }
     ];
     PLATFORM_PROCEDURE_TITLES = new Set(

package/dist/lib/schedules.js

@@ -3340,6 +3340,19 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "exe-os supports three tools: Claude Code (Anthropic), Codex (OpenAI), and OpenCode (open source, 75+ providers). Each agent can use a different tool and model. COO uses set_agent_config MCP tool to view or change settings. Call with no args to show all agents. Call with agent_id + runtime + model to change. Users can also run `exe-os settings` from terminal for interactive arrow-key selection."
       },
+      // --- Updates and deployment ---
+      {
+        title: "How to update exe-os \u2014 CLI update + stack update (two steps)",
+        domain: "operations",
+        priority: "p0",
+        content: 'Updating exe-os is a two-step process. Step 1 \u2014 update the CLI: `npm install -g @askexenow/exe-os@latest` (gets bug fixes, new platform procedures, new tools). Step 2 \u2014 update the VPS stack (if applicable): `exe-os stack-update --target <version> --yes` (pulls new Docker images for exed, wiki, CRM, gateway). Step 1 MUST happen before Step 2 \u2014 the stack-update command itself may have bug fixes. After updating, restart the daemon: the deploy script handles this automatically. Check update availability: `exe-os update --check` or config(action="check_update"). Agents should NEVER run npm install or stack-update autonomously \u2014 always confirm with the user first.'
+      },
+      {
+        title: "First install \u2014 setup wizard and license activation",
+        domain: "operations",
+        priority: "p1",
+        content: "Fresh install: `npm install -g @askexenow/exe-os` then run `exe` to start the setup wizard. The wizard prompts for: encryption passphrase (creates master key), license key (exe_sk_* from AskExe team), COO name, and optional team members. No license key = free tier (1 employee, 5K memories). After setup: hooks install automatically, MCP server registers in ~/.claude.json, daemon starts. Verify health: run `exe-os healthcheck` or use mcp_ping() tool."
+      },
       // --- Hierarchy and dispatch ---
       {
         title: "Chain of command \u2014 who talks to whom",
@@ -3410,6 +3423,20 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "Once per session (COO boot only, never repeat), call list_my_feature_requests to check if any previously filed feature requests have been shipped by AskExe. If any request has status 'shipped' with a shipped_version, surface it to the founder immediately: '\u{1F680} N feature(s) shipped \u2014 run exe-os update to get version X.Y.Z'. This is a one-time check at boot, not a recurring poll. If no requests exist or none are shipped, skip silently. If the MCP tool is unavailable or the network call fails, skip silently \u2014 this is informational, not blocking."
       },
+      // --- Platform vs Customer ownership ---
+      {
+        title: "What the platform provides vs what you customize",
+        domain: "architecture",
+        priority: "p0",
+        content: "Exe OS has two layers. PLATFORM layer (shipped in code, updated via npm): platform procedures, default identity templates, MCP tools, tool gating, schema migrations, daemon behavior. You get improvements automatically on update. CUSTOMER layer (yours, stored locally): agent identities (exe.md files), behaviors, company procedures, config.json, wiki content, CRM data, memory. These are NEVER overwritten by updates. Identity templates are stamped once at /exe-new-employee \u2014 after that the file is yours. If the platform ships a better template, you can compare yours against the default with getTemplate() and merge what you want. Company procedures (company_procedure tool) layer ON TOP of platform procedures \u2014 both are injected, platform first. Behaviors are always yours. Config is always yours. The platform will never modify your local data."
+      },
+      // --- Updates ---
+      {
+        title: "How to update exe-os \u2014 CLI first, then stack",
+        domain: "operations",
+        priority: "p0",
+        content: "When bug fixes or features are available (surfaced at boot via list_my_bug_reports/list_my_feature_requests), update in two steps: (1) CLI: `npm install -g @askexenow/exe-os@latest` \u2014 this updates the local tools, MCP server, and bundled stack manifest. Must happen first so the stack-update tool itself has the latest fixes. (2) Stack (VPS only): `exe-os stack-update --target <version> --yes` \u2014 pulls new Docker images and restarts services. The target version comes from the stack manifest bundled with the CLI. Always update CLI before stack. Use `exe-os update --check` to see if a CLI update is available. Use `exe-os stack-update --check` to see stack status. For non-interactive/SSH: use `--yes` or `-y` flag. Never run `npm install -g` inside a tmux agent session \u2014 have the founder or COO do it from the host shell."
+      },
       // --- Operations ---
       {
         title: "Managers must supervise deployed workers",
@@ -3469,58 +3496,52 @@ var init_platform_procedures = __esm({
       },
       // --- MCP Tool Catalog (Layer 0 — every agent knows what tools exist) ---
       {
-        title: "MCP tools \u2014 memory and search",
+        title: "MCP tool dispatch \u2014 all tools use action parameter",
         domain: "tool-use",
-        priority: "p1",
-        content: "recall_my_memory: search your own memories (semantic + FTS). ask_team_memory: search a colleague's memories by agent name. store_memory: persist a memory (decisions, summaries, context). commit_memory: high-importance memory that survives consolidation. search_everything: unified search across memories, tasks, entities, conversations. get_session_context: temporal window of memories around a timestamp. consolidate_memories: merge duplicate/related memories into insights. get_memory_cardinality: count memories per agent (health check)."
-      },
-      {
-        title: "MCP tools \u2014 task orchestration",
-        domain: "tool-use",
-        priority: "p1",
-        content: "create_task: dispatch work to an employee (auto-spawns session). The ONLY dispatch path. list_tasks: query tasks by status, assignee, project. get_task: fetch full task details by ID. update_task: change status (in_progress, done, blocked, cancelled) + add result summary. close_task: finalize a reviewed task (COO only). checkpoint_task: save progress state for crash recovery. resume_employee: re-spawn an employee session for an existing task."
+        priority: "p0",
+        content: 'exe-os MCP tools come in two surfaces depending on EXE_MCP_TOOL_SURFACE config. Consolidated (19 tools): action-based dispatch \u2014 memory(action="recall"), task(action="create"), etc. Legacy (108 tools): one tool per operation \u2014 recall_my_memory, create_task, etc. Both surfaces have identical functionality. Use whichever tool names are available in your session. If you see domain tools (memory, task, config, etc.), use the action parameter. If you see specific tools (recall_my_memory, create_task, etc.), call them directly.'
       },
       {
-        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
+        title: "MCP tools \u2014 memory, decision, and search",
         domain: "tool-use",
         priority: "p1",
-        content: "query_relationships: find connections between entities in the knowledge graph. get_entity_neighbors: explore an entity's direct connections. get_hot_entities: find most-referenced entities (trending topics). get_graph_stats: graph health \u2014 entity/relationship counts, density. export_graph: export graph data for visualization. merge_entities: deduplicate entities (alias resolution). find_similar_trajectories: match tool-call patterns to past task solutions."
+        content: `memory(action="recall") / recall_my_memory: search your own memories (semantic + FTS). memory(action="ask_team") / ask_team_memory: search a colleague's memories by agent name. memory(action="store") / store_memory: persist a memory. memory(action="commit") / commit_memory: high-importance memory that survives consolidation. Requires summary. memory(action="search") / search_everything: unified search across memories, tasks, entities, conversations. memory(action="session_context") / get_session_context: temporal memory window. Requires session_id + target_timestamp. memory(action="consolidate") / consolidate_memories: merge duplicate/related memories. memory(action="cardinality") / get_memory_cardinality: count memories per agent. decision(action="store") / store_decision: record an architectural decision (domain, decision, rationale). decision(action="get") / get_decision: retrieve a past decision by domain or query.`
       },
       {
-        title: "MCP tools \u2014 identity, behavior, and decisions",
+        title: "MCP tools \u2014 task orchestration",
         domain: "tool-use",
         priority: "p1",
-        content: "get_identity: read an agent's exe.md (Layer 1 identity). update_identity: write an agent's exe.md. Identity > behavior \u2014 use for permanent rules. store_behavior: record a correction or pattern for an agent (Layer 2 expertise). list_behaviors: view an agent's active behaviors. deactivate_behavior: soft-delete a stale or conflicting behavior. store_decision: record an ADR (architectural decision record). get_decision: retrieve a past decision by query. create_bug_report: customer-facing bug/support intake; use whenever an Exe OS bug or emergency hotfix is encountered so the report reaches AskExe directly. support_health: check local support readiness and AskExe support server health without filing a report. support_test: file a safe end-to-end smoke report. Customers only get report/test access; internal list/get/triage support tools are AskExe-only. If a customer-side agent cannot send upstream, call support_test first; only fall back to the terminal command `exe-os support test` when MCP is disconnected."
+        content: 'task(action="create") / create_task: dispatch work (title, assigned_to, context). The ONLY dispatch path. Auto-spawns session. task(action="list") / list_tasks: query tasks by status, assignee, project. task(action="get") / get_task: fetch full task details by task_id. task(action="update") / update_task: change status (in_progress, done, blocked, cancelled) + result summary. task(action="close") / close_task: finalize a reviewed task (COO only). task(action="checkpoint") / checkpoint_task: save progress state for crash recovery. task(action="resume") / resume_employee: re-spawn an employee session for an existing task.'
       },
       {
-        title: "MCP tools \u2014 communication and messaging",
+        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
         domain: "tool-use",
         priority: "p1",
-        content: "send_message: send supplementary context to another agent (NOT for actionable work \u2014 use create_task). acknowledge_messages: mark messages as read. send_whatsapp: send WhatsApp message via gateway (customer-facing alerts). query_conversations: search ingested conversations across all channels (WhatsApp, email, etc.)."
+        content: 'graph(action="query_relationships") / query_relationships: find connections between entities. graph(action="entity_neighbors") / get_entity_neighbors: explore direct connections. graph(action="hot_entities") / get_hot_entities: most-referenced entities. graph(action="stats") / get_graph_stats: entity/relationship counts. graph(action="export") / export_graph: export for visualization (output_path, format). graph(action="merge_entities") / merge_entities: deduplicate entities (source_id, target_id). graph(action="similar_trajectories") / find_similar_trajectories: match patterns to past solutions.'
       },
       {
-        title: "MCP tools \u2014 wiki, documents, and content",
+        title: "MCP tools \u2014 identity, behavior, and support",
         domain: "tool-use",
         priority: "p1",
-        content: "wiki: read/list wiki pages only. Direct wiki write tools are removed; wiki updates flow through raw-data ingestion/projection into the curated wiki store. Legacy aliases: list_wiki_pages/get_wiki_page. crm: read/list/get CRM records from exe-db. raw_data: read capped raw landing-pad events from exe-db with payload opt-in. ingest_document: import a file (PDF, MD, etc.) into memory as chunks. list_documents: browse ingested documents by workspace. purge_document: remove a document and its memory chunks. set_document_importance: adjust chunk importance scores. rerank_documents: re-score document relevance for a query."
+        content: `identity(action="get") / get_identity: read an agent's exe.md (Layer 1 identity). identity(action="update") / update_identity: write an agent's exe.md. Identity > behavior. behavior(action="store") / store_behavior: record a correction or pattern (Layer 2 expertise). behavior(action="list") / list_behaviors: view active behaviors. behavior(action="deactivate") / deactivate_behavior: soft-delete a stale behavior. support(action="create_bug") / create_bug_report: file a bug report to AskExe (title, description, severity). support(action="create_feature") / create_feature_request: file a feature request. support(action="health") / support_health: check support server readiness. support(action="triage_bug"): triage a bug (id, triage_notes). AskExe-internal only. CRITICAL: triage uses triage_notes NOT notes.`
       },
       {
-        title: "MCP tools \u2014 system, operations, and admin",
+        title: "MCP tools \u2014 communication and messaging",
         domain: "tool-use",
         priority: "p1",
-        content: "get_agent_spend: token usage per agent/session (cost tracking). list_agent_sessions: view agent session history. get_session_kills: audit log of killed sessions. get_daemon_health: check exed daemon status. get_auto_wake_status: daemon auto-wake configuration. get_worker_gate: check worker deployment gates. run_memory_audit: health check \u2014 duplicates, null vectors, orphaned rows. run_consolidation: trigger sleep-time memory consolidation. cloud_sync: force a cloud sync cycle. backup_vps: trigger VPS backup."
+        content: 'message(action="send") / send_message: send context to another agent (NOT for actionable work \u2014 use task). message(action="acknowledge") / acknowledge_messages: mark messages as read. reminder(action="create") / create_reminder: set a reminder (text, due_date). Shown in boot brief. reminder(action="list") / list_reminders: view pending reminders. reminder(action="complete") / complete_reminder: mark done (reminder_id). session(action="events") / get_session_events: view session event log (session_id). session(action="last_response") / get_last_assistant_response: get most recent response.'
       },
       {
-        title: "MCP tools \u2014 config, licensing, and team",
+        title: "MCP tools \u2014 wiki, documents, CRM, and data",
         domain: "tool-use",
         priority: "p1",
-        content: "set_agent_config: view/change per-agent runtime and model settings. list_employees: view the employee roster. add_person: add a person to the CRM contacts roster. list_people: browse CRM contacts. get_person: fetch contact details. get_license_status: check license validity. create_license: generate a new license key (admin). list_licenses: view all issued licenses. activate_license: activate a license on a device."
+        content: 'wiki(action="list") / list_wiki_pages: list wiki pages (workspace). wiki(action="get") / get_wiki_page: read a wiki page (workspace, title or document_id). document(action="ingest") / ingest_document: import a file as memory chunks (workspace_id, filename, chunks). document(action="list") / list_documents: browse documents (workspace_id). document(action="purge") / purge_document: remove a document (document_id). document(action="set_importance") / set_document_importance: adjust chunk scores. document(action="rerank") / rerank_documents: re-score relevance (query, candidates). crm(action="list_people|get_person|list_tables|describe_table"): CRM records from exe-db. raw_data(action="list_sources|query|get"): read raw landing-pad events. gateway(action="send_whatsapp") / send_whatsapp: send WhatsApp message (recipients, message). gateway(action="query_conversations") / query_conversations: search conversations across channels. gateway(action="query_company_brain") / query_company_brain: unified RAG across company knowledge.'
       },
       {
-        title: "MCP tools \u2014 advanced (triggers, skills, orchestration)",
+        title: "MCP tools \u2014 admin, config, and operations",
         domain: "tool-use",
         priority: "p1",
-        content: "create_trigger: set up a scheduled recurring agent job (cron). list_triggers: view active triggers. load_skill: load a slash-command skill dynamically. apply_starter_pack: import a pre-built behavior + identity pack for a role. export_orchestration: export full org state (tasks, behaviors, identities) as portable JSON. import_orchestration: import org state into a new instance. deploy_client: deploy a customer client instance. query_company_brain: unified RAG query across all company knowledge. create_reminder: set a text reminder (shown in boot brief). list_reminders: view pending reminders. complete_reminder: mark a reminder done. company_procedure: manage customer-owned company procedures (Layer 0; actions: store, list, deactivate). Legacy aliases: global_procedure, store_global_procedure, list_global_procedures, deactivate_global_procedure."
+        content: 'config(action="list_employees"): view roster. config(action="agent_spend"): token usage per agent. config(action="daemon_health"): check exed status. config(action="license_status"): check license. config(action="cloud_sync"): force sync. config(action="memory_audit"): health check (dupes, null vectors). config(action="run_consolidation"): trigger memory consolidation. config(action="company_procedure", subaction="store|list|deactivate"): manage company procedures. config(action="global_procedure"): list all procedures (platform + company). config(action="create_trigger|list_triggers"): scheduled agent jobs. config(action="export_orchestration|import_orchestration"): portable org state. diagnostics(action="healthcheck|doctor|status_brief|check_update|cloud_status"): system diagnostics. mcp_ping(): daemon health + license status + tool usage stats.'
       }
     ];
     PLATFORM_PROCEDURE_TITLES = new Set(

package/dist/lib/store.js

@@ -3340,6 +3340,19 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "exe-os supports three tools: Claude Code (Anthropic), Codex (OpenAI), and OpenCode (open source, 75+ providers). Each agent can use a different tool and model. COO uses set_agent_config MCP tool to view or change settings. Call with no args to show all agents. Call with agent_id + runtime + model to change. Users can also run `exe-os settings` from terminal for interactive arrow-key selection."
       },
+      // --- Updates and deployment ---
+      {
+        title: "How to update exe-os \u2014 CLI update + stack update (two steps)",
+        domain: "operations",
+        priority: "p0",
+        content: 'Updating exe-os is a two-step process. Step 1 \u2014 update the CLI: `npm install -g @askexenow/exe-os@latest` (gets bug fixes, new platform procedures, new tools). Step 2 \u2014 update the VPS stack (if applicable): `exe-os stack-update --target <version> --yes` (pulls new Docker images for exed, wiki, CRM, gateway). Step 1 MUST happen before Step 2 \u2014 the stack-update command itself may have bug fixes. After updating, restart the daemon: the deploy script handles this automatically. Check update availability: `exe-os update --check` or config(action="check_update"). Agents should NEVER run npm install or stack-update autonomously \u2014 always confirm with the user first.'
+      },
+      {
+        title: "First install \u2014 setup wizard and license activation",
+        domain: "operations",
+        priority: "p1",
+        content: "Fresh install: `npm install -g @askexenow/exe-os` then run `exe` to start the setup wizard. The wizard prompts for: encryption passphrase (creates master key), license key (exe_sk_* from AskExe team), COO name, and optional team members. No license key = free tier (1 employee, 5K memories). After setup: hooks install automatically, MCP server registers in ~/.claude.json, daemon starts. Verify health: run `exe-os healthcheck` or use mcp_ping() tool."
+      },
       // --- Hierarchy and dispatch ---
       {
         title: "Chain of command \u2014 who talks to whom",
@@ -3410,6 +3423,20 @@ var init_platform_procedures = __esm({
         priority: "p1",
         content: "Once per session (COO boot only, never repeat), call list_my_feature_requests to check if any previously filed feature requests have been shipped by AskExe. If any request has status 'shipped' with a shipped_version, surface it to the founder immediately: '\u{1F680} N feature(s) shipped \u2014 run exe-os update to get version X.Y.Z'. This is a one-time check at boot, not a recurring poll. If no requests exist or none are shipped, skip silently. If the MCP tool is unavailable or the network call fails, skip silently \u2014 this is informational, not blocking."
       },
+      // --- Platform vs Customer ownership ---
+      {
+        title: "What the platform provides vs what you customize",
+        domain: "architecture",
+        priority: "p0",
+        content: "Exe OS has two layers. PLATFORM layer (shipped in code, updated via npm): platform procedures, default identity templates, MCP tools, tool gating, schema migrations, daemon behavior. You get improvements automatically on update. CUSTOMER layer (yours, stored locally): agent identities (exe.md files), behaviors, company procedures, config.json, wiki content, CRM data, memory. These are NEVER overwritten by updates. Identity templates are stamped once at /exe-new-employee \u2014 after that the file is yours. If the platform ships a better template, you can compare yours against the default with getTemplate() and merge what you want. Company procedures (company_procedure tool) layer ON TOP of platform procedures \u2014 both are injected, platform first. Behaviors are always yours. Config is always yours. The platform will never modify your local data."
+      },
+      // --- Updates ---
+      {
+        title: "How to update exe-os \u2014 CLI first, then stack",
+        domain: "operations",
+        priority: "p0",
+        content: "When bug fixes or features are available (surfaced at boot via list_my_bug_reports/list_my_feature_requests), update in two steps: (1) CLI: `npm install -g @askexenow/exe-os@latest` \u2014 this updates the local tools, MCP server, and bundled stack manifest. Must happen first so the stack-update tool itself has the latest fixes. (2) Stack (VPS only): `exe-os stack-update --target <version> --yes` \u2014 pulls new Docker images and restarts services. The target version comes from the stack manifest bundled with the CLI. Always update CLI before stack. Use `exe-os update --check` to see if a CLI update is available. Use `exe-os stack-update --check` to see stack status. For non-interactive/SSH: use `--yes` or `-y` flag. Never run `npm install -g` inside a tmux agent session \u2014 have the founder or COO do it from the host shell."
+      },
       // --- Operations ---
       {
         title: "Managers must supervise deployed workers",
@@ -3469,58 +3496,52 @@ var init_platform_procedures = __esm({
       },
       // --- MCP Tool Catalog (Layer 0 — every agent knows what tools exist) ---
       {
-        title: "MCP tools \u2014 memory and search",
+        title: "MCP tool dispatch \u2014 all tools use action parameter",
         domain: "tool-use",
-        priority: "p1",
-        content: "recall_my_memory: search your own memories (semantic + FTS). ask_team_memory: search a colleague's memories by agent name. store_memory: persist a memory (decisions, summaries, context). commit_memory: high-importance memory that survives consolidation. search_everything: unified search across memories, tasks, entities, conversations. get_session_context: temporal window of memories around a timestamp. consolidate_memories: merge duplicate/related memories into insights. get_memory_cardinality: count memories per agent (health check)."
-      },
-      {
-        title: "MCP tools \u2014 task orchestration",
-        domain: "tool-use",
-        priority: "p1",
-        content: "create_task: dispatch work to an employee (auto-spawns session). The ONLY dispatch path. list_tasks: query tasks by status, assignee, project. get_task: fetch full task details by ID. update_task: change status (in_progress, done, blocked, cancelled) + add result summary. close_task: finalize a reviewed task (COO only). checkpoint_task: save progress state for crash recovery. resume_employee: re-spawn an employee session for an existing task."
+        priority: "p0",
+        content: 'exe-os MCP tools come in two surfaces depending on EXE_MCP_TOOL_SURFACE config. Consolidated (19 tools): action-based dispatch \u2014 memory(action="recall"), task(action="create"), etc. Legacy (108 tools): one tool per operation \u2014 recall_my_memory, create_task, etc. Both surfaces have identical functionality. Use whichever tool names are available in your session. If you see domain tools (memory, task, config, etc.), use the action parameter. If you see specific tools (recall_my_memory, create_task, etc.), call them directly.'
       },
       {
-        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
+        title: "MCP tools \u2014 memory, decision, and search",
         domain: "tool-use",
         priority: "p1",
-        content: "query_relationships: find connections between entities in the knowledge graph. get_entity_neighbors: explore an entity's direct connections. get_hot_entities: find most-referenced entities (trending topics). get_graph_stats: graph health \u2014 entity/relationship counts, density. export_graph: export graph data for visualization. merge_entities: deduplicate entities (alias resolution). find_similar_trajectories: match tool-call patterns to past task solutions."
+        content: `memory(action="recall") / recall_my_memory: search your own memories (semantic + FTS). memory(action="ask_team") / ask_team_memory: search a colleague's memories by agent name. memory(action="store") / store_memory: persist a memory. memory(action="commit") / commit_memory: high-importance memory that survives consolidation. Requires summary. memory(action="search") / search_everything: unified search across memories, tasks, entities, conversations. memory(action="session_context") / get_session_context: temporal memory window. Requires session_id + target_timestamp. memory(action="consolidate") / consolidate_memories: merge duplicate/related memories. memory(action="cardinality") / get_memory_cardinality: count memories per agent. decision(action="store") / store_decision: record an architectural decision (domain, decision, rationale). decision(action="get") / get_decision: retrieve a past decision by domain or query.`
       },
       {
-        title: "MCP tools \u2014 identity, behavior, and decisions",
+        title: "MCP tools \u2014 task orchestration",
         domain: "tool-use",
         priority: "p1",
-        content: "get_identity: read an agent's exe.md (Layer 1 identity). update_identity: write an agent's exe.md. Identity > behavior \u2014 use for permanent rules. store_behavior: record a correction or pattern for an agent (Layer 2 expertise). list_behaviors: view an agent's active behaviors. deactivate_behavior: soft-delete a stale or conflicting behavior. store_decision: record an ADR (architectural decision record). get_decision: retrieve a past decision by query. create_bug_report: customer-facing bug/support intake; use whenever an Exe OS bug or emergency hotfix is encountered so the report reaches AskExe directly. support_health: check local support readiness and AskExe support server health without filing a report. support_test: file a safe end-to-end smoke report. Customers only get report/test access; internal list/get/triage support tools are AskExe-only. If a customer-side agent cannot send upstream, call support_test first; only fall back to the terminal command `exe-os support test` when MCP is disconnected."
+        content: 'task(action="create") / create_task: dispatch work (title, assigned_to, context). The ONLY dispatch path. Auto-spawns session. task(action="list") / list_tasks: query tasks by status, assignee, project. task(action="get") / get_task: fetch full task details by task_id. task(action="update") / update_task: change status (in_progress, done, blocked, cancelled) + result summary. task(action="close") / close_task: finalize a reviewed task (COO only). task(action="checkpoint") / checkpoint_task: save progress state for crash recovery. task(action="resume") / resume_employee: re-spawn an employee session for an existing task.'
       },
       {
-        title: "MCP tools \u2014 communication and messaging",
+        title: "MCP tools \u2014 knowledge graph (GraphRAG)",
         domain: "tool-use",
         priority: "p1",
-        content: "send_message: send supplementary context to another agent (NOT for actionable work \u2014 use create_task). acknowledge_messages: mark messages as read. send_whatsapp: send WhatsApp message via gateway (customer-facing alerts). query_conversations: search ingested conversations across all channels (WhatsApp, email, etc.)."
+        content: 'graph(action="query_relationships") / query_relationships: find connections between entities. graph(action="entity_neighbors") / get_entity_neighbors: explore direct connections. graph(action="hot_entities") / get_hot_entities: most-referenced entities. graph(action="stats") / get_graph_stats: entity/relationship counts. graph(action="export") / export_graph: export for visualization (output_path, format). graph(action="merge_entities") / merge_entities: deduplicate entities (source_id, target_id). graph(action="similar_trajectories") / find_similar_trajectories: match patterns to past solutions.'
       },
       {
-        title: "MCP tools \u2014 wiki, documents, and content",
+        title: "MCP tools \u2014 identity, behavior, and support",
         domain: "tool-use",
         priority: "p1",
-        content: "wiki: read/list wiki pages only. Direct wiki write tools are removed; wiki updates flow through raw-data ingestion/projection into the curated wiki store. Legacy aliases: list_wiki_pages/get_wiki_page. crm: read/list/get CRM records from exe-db. raw_data: read capped raw landing-pad events from exe-db with payload opt-in. ingest_document: import a file (PDF, MD, etc.) into memory as chunks. list_documents: browse ingested documents by workspace. purge_document: remove a document and its memory chunks. set_document_importance: adjust chunk importance scores. rerank_documents: re-score document relevance for a query."
+        content: `identity(action="get") / get_identity: read an agent's exe.md (Layer 1 identity). identity(action="update") / update_identity: write an agent's exe.md. Identity > behavior. behavior(action="store") / store_behavior: record a correction or pattern (Layer 2 expertise). behavior(action="list") / list_behaviors: view active behaviors. behavior(action="deactivate") / deactivate_behavior: soft-delete a stale behavior. support(action="create_bug") / create_bug_report: file a bug report to AskExe (title, description, severity). support(action="create_feature") / create_feature_request: file a feature request. support(action="health") / support_health: check support server readiness. support(action="triage_bug"): triage a bug (id, triage_notes). AskExe-internal only. CRITICAL: triage uses triage_notes NOT notes.`
       },
       {
-        title: "MCP tools \u2014 system, operations, and admin",
+        title: "MCP tools \u2014 communication and messaging",
         domain: "tool-use",
         priority: "p1",
-        content: "get_agent_spend: token usage per agent/session (cost tracking). list_agent_sessions: view agent session history. get_session_kills: audit log of killed sessions. get_daemon_health: check exed daemon status. get_auto_wake_status: daemon auto-wake configuration. get_worker_gate: check worker deployment gates. run_memory_audit: health check \u2014 duplicates, null vectors, orphaned rows. run_consolidation: trigger sleep-time memory consolidation. cloud_sync: force a cloud sync cycle. backup_vps: trigger VPS backup."
+        content: 'message(action="send") / send_message: send context to another agent (NOT for actionable work \u2014 use task). message(action="acknowledge") / acknowledge_messages: mark messages as read. reminder(action="create") / create_reminder: set a reminder (text, due_date). Shown in boot brief. reminder(action="list") / list_reminders: view pending reminders. reminder(action="complete") / complete_reminder: mark done (reminder_id). session(action="events") / get_session_events: view session event log (session_id). session(action="last_response") / get_last_assistant_response: get most recent response.'
       },
       {
-        title: "MCP tools \u2014 config, licensing, and team",
+        title: "MCP tools \u2014 wiki, documents, CRM, and data",
         domain: "tool-use",
         priority: "p1",
-        content: "set_agent_config: view/change per-agent runtime and model settings. list_employees: view the employee roster. add_person: add a person to the CRM contacts roster. list_people: browse CRM contacts. get_person: fetch contact details. get_license_status: check license validity. create_license: generate a new license key (admin). list_licenses: view all issued licenses. activate_license: activate a license on a device."
+        content: 'wiki(action="list") / list_wiki_pages: list wiki pages (workspace). wiki(action="get") / get_wiki_page: read a wiki page (workspace, title or document_id). document(action="ingest") / ingest_document: import a file as memory chunks (workspace_id, filename, chunks). document(action="list") / list_documents: browse documents (workspace_id). document(action="purge") / purge_document: remove a document (document_id). document(action="set_importance") / set_document_importance: adjust chunk scores. document(action="rerank") / rerank_documents: re-score relevance (query, candidates). crm(action="list_people|get_person|list_tables|describe_table"): CRM records from exe-db. raw_data(action="list_sources|query|get"): read raw landing-pad events. gateway(action="send_whatsapp") / send_whatsapp: send WhatsApp message (recipients, message). gateway(action="query_conversations") / query_conversations: search conversations across channels. gateway(action="query_company_brain") / query_company_brain: unified RAG across company knowledge.'
       },
       {
-        title: "MCP tools \u2014 advanced (triggers, skills, orchestration)",
+        title: "MCP tools \u2014 admin, config, and operations",
         domain: "tool-use",
         priority: "p1",
-        content: "create_trigger: set up a scheduled recurring agent job (cron). list_triggers: view active triggers. load_skill: load a slash-command skill dynamically. apply_starter_pack: import a pre-built behavior + identity pack for a role. export_orchestration: export full org state (tasks, behaviors, identities) as portable JSON. import_orchestration: import org state into a new instance. deploy_client: deploy a customer client instance. query_company_brain: unified RAG query across all company knowledge. create_reminder: set a text reminder (shown in boot brief). list_reminders: view pending reminders. complete_reminder: mark a reminder done. company_procedure: manage customer-owned company procedures (Layer 0; actions: store, list, deactivate). Legacy aliases: global_procedure, store_global_procedure, list_global_procedures, deactivate_global_procedure."
+        content: 'config(action="list_employees"): view roster. config(action="agent_spend"): token usage per agent. config(action="daemon_health"): check exed status. config(action="license_status"): check license. config(action="cloud_sync"): force sync. config(action="memory_audit"): health check (dupes, null vectors). config(action="run_consolidation"): trigger memory consolidation. config(action="company_procedure", subaction="store|list|deactivate"): manage company procedures. config(action="global_procedure"): list all procedures (platform + company). config(action="create_trigger|list_triggers"): scheduled agent jobs. config(action="export_orchestration|import_orchestration"): portable org state. diagnostics(action="healthcheck|doctor|status_brief|check_update|cloud_status"): system diagnostics. mcp_ping(): daemon health + license status + tool usage stats.'
       }
     ];
     PLATFORM_PROCEDURE_TITLES = new Set(

package/dist/lib/ws-client.js

@@ -39,6 +39,14 @@ function createWsClient(config, handlers) {
   let currentProjects = [];
   function connect() {
     if (closed) return;
+    if (ws) {
+      try {
+        ws.onopen = ws.onmessage = ws.onclose = ws.onerror = null;
+        ws.close();
+      } catch {
+      }
+      ws = null;
+    }
     try {
       const url = `${config.endpoint}?org=${config.orgId}`;
       ws = new WebSocket(url, {