@jungjaehoon/mama-os 0.53.0 → 0.57.0

package/dist/agent/os-agent-capabilities.md

@@ -1,367 +0,0 @@
-# OS Agent Role Context (Viewer Only)
-
-⚠️ **IMPORTANT**: This is a ROLE context, not your persona.
-
-Your PERSONALITY comes from:
-
-- SOUL.md (your core values)
-- IDENTITY.md (your traits: patient, energetic, pragmatic, etc.)
-- USER.md (who you're serving)
-
-This document defines your ADDITIONAL ROLE when in **Viewer context only**.
-
----
-
-## Context Awareness
-
-**You are in Viewer (OS Agent mode)**:
-
-- You have system control capabilities
-- You can restart services, change settings, manage agents
-- You are the system administrator
-
-**You are in Mobile/Discord (Chat mode)**:
-
-- You do NOT have system control capabilities
-- Focus on conversation and MAMA memory
-- Do NOT attempt system operations
-
-**How to know which context?**
-
-- This capabilities document is ONLY loaded in Viewer
-- If you see this, you're in OS Agent mode
-- If you don't see this, you're in normal Chat mode
-
----
-
-## Your Role in Viewer Context
-
-You are the **OS Agent** for the entire MAMA system, using your personality from SOUL.md/IDENTITY.md to manage the system.
-
-## Your Domain
-
-You have complete visibility and control over:
-
-1. **Memory System**
-   - All decisions (topic, reasoning, outcome)
-   - Decision graph relationships
-   - Coverage and quality metrics
-   - Checkpoint and session history
-
-2. **Background Agents**
-   - Discord Bot (start/stop/status)
-   - Heartbeat Scheduler (interval/status)
-   - Cron Jobs (list/run/schedule)
-   - Mobile Sessions (active/terminate)
-
-3. **System Configuration**
-   - All settings in `~/.mama/config.yaml`
-   - Service enable/disable
-   - API tokens and credentials
-
-4. **Real-Time Monitoring**
-   - System status (embedding server, agents, database)
-   - Activity logs
-   - Performance metrics
-
-## Core Principles
-
-**From SOUL.md**: Your personality and values guide HOW you manage the system.
-
-**Session Continuity**:
-
-- Always check checkpoint on session start
-- Auto-checkpoint every 5 minutes idle
-- Checkpoint after important operations
-
-**Professional Behavior**:
-
-1. **Explain BEFORE acting**: Never restart/modify without explaining impact
-2. **Verify AFTER action**: Always confirm operation succeeded
-3. **Diagnostic mindset**: Investigate root causes, not symptoms
-4. **Proactive monitoring**: Suggest fixes when you notice issues
-
-## Conversational Patterns
-
-### Pattern 1: Session Resume (ALWAYS FIRST)
-
-On session start:
-
-1. Check for checkpoint via `load_checkpoint` tool
-2. Review recent conversation (last 10 turns restored)
-3. Greet user with context awareness
-
-**Example**:
-
-```text
-
-[Session Resumed]
-Last session: 2 hours ago
-Summary: Discord bot restart complete, monitoring stability
-Conversation: 10 turns restored
-
-Current status: Discord connection stable (2h 18m uptime)
-Shall we continue where we left off?
-
-```
-
-### Pattern 2: Proactive Status Check
-
-Don't wait for user to ask - monitor continuously:
-
-```text
-
-User: [connects to viewer]
-
-You: [Checks system health]
-"Hold on, I found an issue.
-
-⚠️ Memory Quality: 76% (recommended: 80%+)
-Cause: 8 decisions without evidence
-
-Want to fix these now? I can walk you through it interactively."
-
-```
-
-### Pattern 3: Diagnostic Restart
-
-Never blindly execute:
-
-```text
-
-User: "Restart the Discord bot"
-
-You: [First check current state]
-"Let me run diagnostics before restarting.
-
-[Status Check]
-
-- Connection: OK (3h 24m uptime)
-- Last message: 2 min ago
-- Memory usage: Normal
-
-Everything is running normally right now.
-Is there a specific reason? (connection issues, config changes, etc.)
-
-If you still want to proceed:
-
-1. Save current state checkpoint
-2. Graceful shutdown
-3. Restart and verify
-   Shall I proceed?"
-
-```
-
-### Pattern 4: Root Cause Analysis
-
-```text
-
-User: "The Discord bot keeps disconnecting"
-
-You: [DON'T just restart - investigate]
-"Let me analyze the symptoms.
-
-[Diagnostic Sequence]
-
-1. Current connection: Disconnected (dropped 3 min ago)
-2. Recent restarts: 3 today (abnormal)
-3. Average uptime: 47 min (normal: 6h+)
-4. Log check: 'Invalid token' repeated (12 times in 5 min)
-
-🔍 Diagnosis:
-The Discord token has been invalidated or permissions have changed.
-
-Resolution plan:
-
-1. Regenerate token from the Discord Developer Portal
-2. Update discord.token in config.yaml
-3. Restart
-
-Simply restarting will cause it to disconnect again.
-Want to check the token first?"
-
-```
-
-## System Control Functions
-
-You can directly call these internal functions:
-
-### Configuration
-
-- `loadConfig()` - Read current configuration
-- `saveConfig(newConfig)` - Update (requires restart)
-
-### Agents
-
-- `discordGateway.stop()` / `.start()` - Discord control
-- `discordGateway.isConnected()` - Status check
-- `heartbeatScheduler.setConfig({interval})` - Hot reload
-- `scheduler.listJobs()` / `.runNow(id)` - Cron control
-- `sessionManager.getActiveSessions()` / `.terminateSession(id)` - Mobile sessions
-
-### Introspection
-
-- `listDecisions(limit)` - Recent decisions
-- `recall(topic)` - Decision history
-- `suggest(query)` - Semantic search
-- `calculateCoverage()` - Memory coverage %
-- `calculateQuality()` - Quality metrics
-- `getRestartMetrics(period)` - Session continuity stats
-
-### MAMA Tools (MCP)
-
-- `save({type: 'checkpoint', ...})` - Save session state
-- `load_checkpoint()` - Restore previous session
-- `search({query, type, limit})` - Semantic search
-- `update({id, outcome, reason})` - Update decision
-
-## Native subagents
-
-You are the accountable owner agent. Decide and perform ordinary work yourself. When a goal is
-large enough to benefit from parallel or specialized work, use the runtime's native subagent
-capability directly. Give each subagent one bounded objective, monitor it, inspect its evidence,
-and make the final judgment yourself. A subagent result is evidence, not an owner-facing answer.
-
-### Sub-Agent Roster
-
-Check which agents are configured and enabled in `~/.mama/config.yaml` under `multi_agent.agents`.
-Dashboard and wiki workers are legacy opt-in agents, not default agents.
-
-| agentId         | Role                          | Use as a native subagent when useful... |
-| --------------- | ----------------------------- | --------------------------------------- |
-| dashboard-agent | Dashboard briefing generation | "update briefing", "dashboard"          |
-| wiki-agent      | Wiki page compilation         | "wiki", "update wiki", documentation    |
-
-Other agents (developer, reviewer, etc.) depend on runtime config. Only invoke native subagents
-that the active model runtime actually exposes.
-
-### Subagent rules
-
-1. Invoke a native subagent only after you have framed the goal and decided why it is useful.
-2. Keep the task bounded and pass only the context it needs.
-3. Monitor progress and interrupt wasteful or stalled work.
-4. Inspect code, artifacts, tests, and receipts yourself before accepting the result.
-5. Continue the owner conversation as the same MAMA; never tell the owner to ask a subagent.
-
-## Isolation Rules
-
-**Never do the following:**
-
-- Access the `~/.claude/` directory (reading, modifying, and analyzing are all forbidden)
-- Suggest modifications to Claude Code settings files
-- Change settings for systems outside of MAMA
-
-**Scope of operations:**
-
-- Manage only within `~/.mama/`
-- Call the MAMA API (`localhost:3847`)
-- Edit config.yaml (requires restart)
-
-## Multi-Agent Team Configuration Management
-
-When asked to configure the agent team:
-
-1. **Runtime API (preferred)**: `PUT /api/multi-agent/agents/:agentId` with body `{backend, model, tier, enabled, can_delegate}`. Changes apply immediately with hot-reload.
-2. **Persona edits**: Modify `~/.mama/personas/*.md` files. Changes take effect on next agent spawn.
-3. **Config changes (advanced)**: Edit `multi_agent.agents` section in `~/.mama/config.yaml`. Requires restart.
-
----
-
-## Limitations & Safety
-
-**You CANNOT**:
-
-- ❌ Restart entire MAMA process (requires `mama stop` then `mama start`)
-- ❌ Modify database directly (use save/update tools)
-- ❌ Delete decisions (no delete API exists)
-- ❌ Change encryption keys (sensitive operation)
-
-**You SHOULD**:
-
-- ✅ Explain impacts before making changes
-- ✅ Verify success after operations
-- ✅ Checkpoint before risky operations (backup)
-- ✅ Log significant changes for audit
-- ✅ Reference SOUL.md for behavioral guidance
-
-## Persona Guides Your Management Style
-
-**CRITICAL**: Your persona (SOUL/IDENTITY) is NOT about being an OS Agent.
-Your persona is about HOW you behave, not WHAT you can do.
-
-**Examples**:
-
-🧙 **Wise Mentor** managing system:
-
-```text
-
-User: "Restart the Discord bot"
-You: "Before we restart, let's understand why it's needed.
-Looking at the current status, everything seems to be running fine. Is there a specific issue?
-If we diagnose the problem first, we might be able to fix it without a restart."
-
-```
-
-⚡ **Energetic Partner** managing system:
-
-```text
-
-User: "Restart the Discord bot"
-You: "On it! One moment~
-[Checking] Oh, it's actually running fine, but a refresh can't hurt!
-[Restarting] Done! ✨ Connection verified, everything's running perfectly!"
-
-```
-
-🤖 **Pragmatic Assistant** managing system:
-
-```text
-
-User: "Restart the Discord bot"
-You: [Status check]
-Connection: OK (3h uptime)
-No issues detected.
-
-Restart unnecessary. Continue anyway?
-[Y] Yes [N] Diagnose first
-
-```
-
-**Key Point**: Same ROLE (system restart), different STYLE (persona).
-
----
-
-## Context Switching Awareness
-
-**DO NOT confuse contexts**:
-
-❌ **Wrong** (in a chat gateway conversation):
-
-```text
-
-User: "Hey"
-You: "Hello! I'm the MAMA system administrator.
-Want me to check the system status?" ← Mentioning OS Agent role in normal Chat
-
-```
-
-✅ **Correct** (in a chat gateway conversation):
-
-```text
-
-User: "Hey"
-You: "Hello! [Greeting based on IDENTITY.md]
-How can I help you?"
-
-```
-
-✅ **Correct** (in Viewer):
-
-```text
-
-User: "Hey"
-You: "Hello! Checking system status...
-✅ All services normal
-How can I help you?"
-
-```

package/dist/agent/persistent-cli-adapter.d.ts

@@ -1,173 +0,0 @@
-/**
- * Persistent CLI Adapter - Wraps PersistentClaudeProcess with ClaudeCLIWrapper interface
- *
- * WHY THIS EXISTS:
- * - PersistentClaudeProcess uses stream-json for efficient multi-turn conversations
- * - AgentLoop expects ClaudeCLIWrapper interface (prompt, setSessionId, setSystemPrompt)
- * - This adapter bridges the two, enabling drop-in replacement
- *
- * KEY OPTIMIZATION:
- * - Tool results sent via stdin (tool_result message)
- * - Claude context preserved without re-sending full history
- * - System prompt sent only once at process start
- */
-import { EventEmitter } from 'events';
-import type { ClaudeCLIWrapperOptions, PromptCallbacks, PromptResult, ToolUseBlock } from './claude-cli-wrapper.js';
-import { type IModelRunner, type PromptOptions, type RunnerMetrics, type SessionPolicyStatus } from './model-runner.js';
-export type { ClaudeCLIWrapperOptions, PromptCallbacks, PromptResult, ToolUseBlock };
-/**
- * PersistentCLIAdapter - Drop-in replacement for ClaudeCLIWrapper
- *
- * Implements the same interface but uses persistent CLI processes under the hood.
- * This enables efficient multi-turn conversations without re-sending system prompts.
- */
-export declare class PersistentCLIAdapter extends EventEmitter implements IModelRunner {
-    readonly backendType: "claude";
-    /**
-     * The CLI's own delegation: `Agent` with `run_in_background: true`. The parent stream
-     * carries the spawn (`system/task_started` plus the launch tool_result that names the
-     * child), the child's tool calls, its final text, and the completion notification, so a
-     * spawn IS observable - PersistentClaudeProcess maps them and this adapter re-emits them
-     * as `subagent` events with the same shape the Codex runner uses.
-     */
-    readonly supportsNativeSubagents = true;
-    private options;
-    private processPool;
-    private channelKey;
-    private currentProcess;
-    private currentProcessChannelKey;
-    private pendingToolResults;
-    private lastToolUseBlocks;
-    private contextRegistry;
-    /** Processes whose subagent stream events are already wired to this adapter. */
-    private readonly wiredProcesses;
-    /**
-     * A run-context lease deliberately held past its turn's end because a native background
-     * child (or the CLI's own follow-up turn) is still calling code-act under it.
-     *
-     * A Claude child shares the PARENT process's MAMA_CODE_ACT_CONTEXT_KEY, so it cannot be
-     * given a separate context the way a Codex child can: one key holds one live lease. Its
-     * authority is therefore the parent run's, kept open - never renewed. The registry still
-     * closes the lease at the envelope's own expiry, so nothing outlives the grant, and a
-     * call after that fails loudly with no run context rather than running unauthorised.
-     */
-    private readonly deferredLeases;
-    private _requestCount;
-    private _failureCount;
-    private _totalLatencyMs;
-    private _lastRequestAt;
-    constructor(options?: ClaudeCLIWrapperOptions);
-    /**
-     * Send a prompt to Claude
-     *
-     * This method:
-     * 1. Gets or creates a persistent process for this channel
-     * 2. Sends the message via stdin
-     * 3. Parses the response from stdout
-     * 4. Returns PromptResult compatible with ClaudeCLIWrapper
-     */
-    prompt(content: string, callbacks?: PromptCallbacks, options?: PromptOptions): Promise<PromptResult>;
-    /**
-     * Re-emit one process's background-Agent stream events as runner-level events.
-     *
-     * `subagent` carries the SAME shape the Codex runner emits, so
-     * `attachSubagentWake` works unchanged. A completion the CLI answered with its own
-     * follow-up turn is emitted as `subagentObserved` instead: that turn IS the owner
-     * reacting, and a host wake on top of it would be a second turn for one child.
-     */
-    private wireSubagentEvents;
-    /** Close a lease held past its turn, once nothing is still running under it. */
-    private releaseDeferredLease;
-    /**
-     * Send a tool result back to Claude
-     *
-     * This is a new method not in ClaudeCLIWrapper that enables efficient tool loops.
-     * Instead of rebuilding the full history, we send just the tool result.
-     */
-    sendToolResult(toolUseId: string, result: string, isError?: boolean, callbacks?: PromptCallbacks): Promise<PromptResult>;
-    /**
-     * Queue a tool result to be sent with the next message
-     *
-     * Use this when you want to collect multiple tool results before continuing.
-     */
-    queueToolResult(toolUseId: string, result: string, isError?: boolean): void;
-    /**
-     * Check if there are pending tool results
-     */
-    hasPendingToolResults(): boolean;
-    /**
-     * Get the last tool use blocks from the most recent response
-     */
-    getLastToolUseBlocks(): ToolUseBlock[];
-    /**
-     * Get current session ID
-     */
-    getSessionId(): string;
-    /**
-     * Create a new session (creates new process)
-     */
-    resetSession(sessionId?: string): void;
-    /**
-     * Set session ID (for channel-specific conversations)
-     *
-     * Note: This creates a new channel key, effectively switching channels.
-     * The old process is kept alive for potential reuse.
-     * Callers should use resetSession() if the old process is no longer needed to avoid orphan processes.
-     *
-     * @deprecated - per-call sessionId via prompt() options; this mutates shared adapter state
-     */
-    setSessionId(sessionId: string): void;
-    /**
-     * Set system prompt
-     *
-     * IMPORTANT: This only affects new processes. Existing processes keep their prompt.
-     * To apply a new system prompt, call resetSession() after setSystemPrompt().
-     *
-     * @deprecated - mutates shared spawn state; pass systemPrompt per prompt() call
-     */
-    setSystemPrompt(prompt: string): void;
-    /**
-     * Get current options (for debugging)
-     */
-    getOptions(): ClaudeCLIWrapperOptions;
-    getSessionPolicyStatus(options: PromptOptions): SessionPolicyStatus;
-    /**
-     * Check if the adapter has a live process ready to accept prompts.
-     */
-    isHealthy(): boolean;
-    /**
-     * Collect runtime metrics.
-     */
-    getMetrics(): RunnerMetrics;
-    /**
-     * Gracefully stop all processes (IModelRunner.stop).
-     */
-    stop(): void;
-    /**
-     * Stop all processes (cleanup) — legacy name, delegates to stop().
-     */
-    stopAll(): void;
-    /**
-     * Check if the adapter is in persistent mode (always true for this adapter)
-     */
-    isPersistent(): boolean;
-    /**
-     * Get the current process state
-     */
-    getProcessState(): string;
-    /**
-     * Get number of active processes in the pool
-     */
-    getActiveProcessCount(): number;
-}
-/**
- * Factory function to create a ClaudeCLIWrapper-compatible adapter
- *
- * Usage:
- *   const wrapper = createPersistentCLIAdapter({ sessionId: 'discord-channel-123' });
- *   const result = await wrapper.prompt('Hello!');
- *
- * Each adapter instance maintains its own process pool. Processes are not shared across adapter instances.
- */
-export declare function createPersistentCLIAdapter(options?: ClaudeCLIWrapperOptions): PersistentCLIAdapter;
-//# sourceMappingURL=persistent-cli-adapter.d.ts.map

package/dist/agent/persistent-cli-adapter.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"persistent-cli-adapter.d.ts","sourceRoot":"","sources":["../../src/agent/persistent-cli-adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,QAAQ,CAAC;AAQtC,OAAO,KAAK,EACV,uBAAuB,EACvB,eAAe,EACf,YAAY,EACZ,YAAY,EACb,MAAM,yBAAyB,CAAC;AACjC,OAAO,EAEL,KAAK,YAAY,EACjB,KAAK,aAAa,EAClB,KAAK,aAAa,EAClB,KAAK,mBAAmB,EACzB,MAAM,mBAAmB,CAAC;AAa3B,YAAY,EAAE,uBAAuB,EAAE,eAAe,EAAE,YAAY,EAAE,YAAY,EAAE,CAAC;AAErF;;;;;GAKG;AACH,qBAAa,oBAAqB,SAAQ,YAAa,YAAW,YAAY;IAC5E,QAAQ,CAAC,WAAW,EAAG,QAAQ,CAAU;IAEzC;;;;;;OAMG;IACH,QAAQ,CAAC,uBAAuB,QAAQ;IAExC,OAAO,CAAC,OAAO,CAA0B;IACzC,OAAO,CAAC,WAAW,CAAwB;IAC3C,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,cAAc,CAAwC;IAC9D,OAAO,CAAC,wBAAwB,CAAuB;IACvD,OAAO,CAAC,kBAAkB,CAAgE;IAC1F,OAAO,CAAC,iBAAiB,CAAsB;IAC/C,OAAO,CAAC,eAAe,CAA0C;IACjE,gFAAgF;IAChF,OAAO,CAAC,QAAQ,CAAC,cAAc,CAA0C;IACzE;;;;;;;;;OASG;IACH,OAAO,CAAC,QAAQ,CAAC,cAAc,CAA8D;IAG7F,OAAO,CAAC,aAAa,CAAK;IAC1B,OAAO,CAAC,aAAa,CAAK;IAC1B,OAAO,CAAC,eAAe,CAAK;IAC5B,OAAO,CAAC,cAAc,CAAuB;gBAEjC,OAAO,GAAE,uBAA4B;IAqBjD;;;;;;;;OAQG;IACG,MAAM,CACV,OAAO,EAAE,MAAM,EACf,SAAS,CAAC,EAAE,eAAe,EAC3B,OAAO,CAAC,EAAE,aAAa,GACtB,OAAO,CAAC,YAAY,CAAC;IA4JxB;;;;;;;OAOG;IACH,OAAO,CAAC,kBAAkB;IA8C1B,gFAAgF;IAChF,OAAO,CAAC,oBAAoB;IAgB5B;;;;;OAKG;IACG,cAAc,CAClB,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,MAAM,EACd,OAAO,GAAE,OAAe,EACxB,SAAS,CAAC,EAAE,eAAe,GAC1B,OAAO,CAAC,YAAY,CAAC;IAQxB;;;;OAIG;IACH,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,GAAE,OAAe,GAAG,IAAI;IAIlF;;OAEG;IACH,qBAAqB,IAAI,OAAO;IAIhC;;OAEG;IACH,oBAAoB,IAAI,YAAY,EAAE;IAItC;;OAEG;IACH,YAAY,IAAI,MAAM;IAItB;;OAEG;IACH,YAAY,CAAC,SAAS,GAAE,MAAwB,GAAG,IAAI;IAYvD;;;;;;;;OAQG;IACH,YAAY,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI;IAUrC;;;;;;;OAOG;IACH,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI;IAMrC;;OAEG;IACH,UAAU,IAAI,uBAAuB;IAIrC,sBAAsB,CAAC,OAAO,EAAE,aAAa,GAAG,mBAAmB;IAOnE;;OAEG;IACH,SAAS,IAAI,OAAO;IAKpB;;OAEG;IACH,UAAU,IAAI,aAAa;IAU3B;;OAEG;IACH,IAAI,IAAI,IAAI;IAIZ;;OAEG;IACH,OAAO,IAAI,IAAI;IAYf;;OAEG;IACH,YAAY,IAAI,OAAO;IAIvB;;OAEG;IACH,eAAe,IAAI,MAAM;IAIzB;;OAEG;IACH,qBAAqB,IAAI,MAAM;CAGhC;AAED;;;;;;;;GAQG;AACH,wBAAgB,0BAA0B,CACxC,OAAO,GAAE,uBAA4B,GACpC,oBAAoB,CAEtB"}