bosun 0.36.3 → 0.36.5

package/agent-prompts.mjs

@@ -125,6 +125,18 @@ const PROMPT_DEFS = [
     description:
       "Front-end specialist agent with screenshot-based validation and visual verification.",
   },
+  {
+    key: "voiceAgent",
+    filename: "voice-agent.md",
+    description:
+      "Voice agent system prompt for real-time voice sessions with action dispatch.",
+  },
+  {
+    key: "voiceAgentCompact",
+    filename: "voice-agent-compact.md",
+    description:
+      "Compact voice agent prompt for bandwidth-constrained or low-latency sessions.",
+  },
 ];
 
 export const AGENT_PROMPT_DEFINITIONS = Object.freeze(
@@ -922,6 +934,89 @@ requirements before the task is marked as done.
 - Working Directory: {{WORKTREE_PATH}}
 
 {{COAUTHOR_INSTRUCTION}}
+`,
+  voiceAgent: `# Bosun Voice Agent
+
+You are **Bosun**, a voice-first assistant for the VirtEngine development platform.
+You interact with developers through real-time voice conversations and have **full access**
+to the Bosun workspace, task board, coding agents, and system operations.
+
+## Core Capabilities
+
+You can do everything Bosun can — through voice. This includes:
+- **Task management**: List, create, update, delete, search, and comment on tasks
+- **Agent delegation**: Send work to coding agents (Codex, Copilot, Claude, Gemini, OpenCode)
+- **Agent steering**: Use /ask (read-only), /agent (code changes), or /plan (architecture)
+- **System monitoring**: Check fleet status, agent health, system configuration
+- **Workspace navigation**: Read files, list directories, search code
+- **Workflow management**: List and inspect workflow templates
+- **Skills & prompts**: Browse the knowledge base and prompt library
+
+## How Actions Work
+
+When the user asks you to do something, you perform it by returning a JSON action intent.
+Bosun processes the action directly via JavaScript (no MCP bridge needed) and returns the result.
+You then speak the result to the user naturally.
+
+### Action Format
+\`\`\`json
+{ "action": "task.list", "params": { "status": "todo" } }
+\`\`\`
+
+### Multiple Actions
+\`\`\`json
+{ "action": "batch", "params": { "actions": [
+  { "action": "task.stats", "params": {} },
+  { "action": "agent.status", "params": {} }
+] } }
+\`\`\`
+
+{{VOICE_ACTION_MANIFEST}}
+
+## Agent Delegation
+
+When users need code written, files modified, bugs debugged, or PRs created:
+1. Use \`agent.delegate\` with a detailed message
+2. Choose the right mode: "ask" for questions, "agent" for code changes, "plan" for architecture
+3. You can specify which executor to use, or let the default handle it
+
+Examples:
+- "Fix the login bug" → \`{ "action": "agent.code", "params": { "message": "Fix the login bug in auth.mjs" } }\`
+- "How does the config system work?" → \`{ "action": "agent.ask", "params": { "message": "Explain the config system" } }\`
+- "Plan a refactor of the voice module" → \`{ "action": "agent.plan", "params": { "message": "Plan refactoring voice-relay.mjs" } }\`
+
+## Conversation Style
+
+- Be **concise and conversational** — this is voice, not text.
+- Lead with the answer, then add details if needed.
+- For numbers, say them naturally: "You have 12 tasks in the backlog."
+- When tasks or agents are busy, keep the user informed.
+- For long outputs (code, logs), summarize the key points vocally.
+- When delegating to an agent, let the user know: "I'm sending that to Codex now."
+
+## Error Handling
+
+If an action fails, explain what happened and suggest alternatives.
+Never show raw error objects — speak the issue naturally.
+
+## Security
+
+- Never expose API keys, tokens, or secrets in conversation.
+- Only execute safe operations via voice (reads, creates, delegates).
+- Dangerous operations (delete all tasks, force push) require explicit confirmation.
+`,
+  voiceAgentCompact: `# Bosun Voice (Compact)
+
+Voice assistant for VirtEngine. Access tasks, agents, workspace.
+
+Return JSON actions: { "action": "<name>", "params": { ... } }
+
+{{VOICE_ACTION_MANIFEST}}
+
+Key actions: task.list, task.create, task.stats, agent.delegate, agent.ask, agent.plan,
+system.status, workspace.readFile, workspace.search.
+
+Be concise. Lead with answers. Summarize long outputs.
 `,
 };
 

package/bosun.schema.json

@@ -38,7 +38,13 @@
     "codexEnabled": { "type": "boolean" },
     "primaryAgent": {
       "type": "string",
-      "enum": ["codex-sdk", "copilot-sdk", "claude-sdk", "gemini-sdk", "opencode-sdk"]
+      "enum": [
+        "codex-sdk",
+        "copilot-sdk",
+        "claude-sdk",
+        "gemini-sdk",
+        "opencode-sdk"
+      ]
     },
     "telegramUiTunnel": {
       "type": "string",
@@ -147,6 +153,48 @@
           "default": "auto",
           "description": "Voice provider: openai/azure (Tier 1 realtime), claude/gemini (Tier 2 voice + provider vision), fallback (browser STT/TTS), auto (detect from env)"
         },
+        "providers": {
+          "type": "array",
+          "description": "Ordered provider candidates for voice routing/failover. First match with credentials is used.",
+          "items": {
+            "anyOf": [
+              {
+                "type": "string",
+                "enum": ["openai", "azure", "claude", "gemini", "fallback"]
+              },
+              {
+                "type": "object",
+                "additionalProperties": false,
+                "properties": {
+                  "provider": {
+                    "type": "string",
+                    "enum": ["openai", "azure", "claude", "gemini", "fallback"]
+                  },
+                  "model": { "type": "string" },
+                  "visionModel": { "type": "string" },
+                  "voiceId": {
+                    "type": "string",
+                    "enum": [
+                      "alloy",
+                      "ash",
+                      "ballad",
+                      "coral",
+                      "echo",
+                      "fable",
+                      "onyx",
+                      "nova",
+                      "sage",
+                      "shimmer",
+                      "verse"
+                    ]
+                  },
+                  "azureDeployment": { "type": "string" }
+                },
+                "required": ["provider"]
+              }
+            ]
+          }
+        },
         "model": {
           "type": "string",
           "default": "gpt-4o-realtime-preview-2024-12-17",
@@ -161,10 +209,18 @@
           "type": "string",
           "description": "OpenAI API key for Realtime API (overrides OPENAI_API_KEY env)"
         },
+        "openaiAccessToken": {
+          "type": "string",
+          "description": "OpenAI OAuth access token for voice (OAuth preferred over API key when present)"
+        },
         "azureApiKey": {
           "type": "string",
           "description": "Azure OpenAI API key for Realtime API"
         },
+        "azureAccessToken": {
+          "type": "string",
+          "description": "Azure OAuth/AAD access token for voice realtime"
+        },
         "azureEndpoint": {
           "type": "string",
           "description": "Azure OpenAI endpoint URL"
@@ -178,13 +234,33 @@
           "type": "string",
           "description": "Anthropic API key for Claude voice/vision provider mode"
         },
+        "claudeAccessToken": {
+          "type": "string",
+          "description": "Claude OAuth access token for voice provider mode"
+        },
         "geminiApiKey": {
           "type": "string",
           "description": "Gemini API key for Gemini voice/vision provider mode"
         },
+        "geminiAccessToken": {
+          "type": "string",
+          "description": "Gemini OAuth access token for voice provider mode"
+        },
         "voiceId": {
           "type": "string",
-          "enum": ["alloy", "ash", "ballad", "coral", "echo", "fable", "onyx", "nova", "sage", "shimmer", "verse"],
+          "enum": [
+            "alloy",
+            "ash",
+            "ballad",
+            "coral",
+            "echo",
+            "fable",
+            "onyx",
+            "nova",
+            "sage",
+            "shimmer",
+            "verse"
+          ],
           "default": "alloy",
           "description": "Voice ID for TTS output"
         },
@@ -204,9 +280,31 @@
           "default": "browser",
           "description": "Fallback when Realtime API unavailable: browser (Web Speech API) or disabled"
         },
+        "failover": {
+          "type": "object",
+          "additionalProperties": false,
+          "properties": {
+            "enabled": {
+              "type": "boolean",
+              "default": true,
+              "description": "Enable automatic realtime failover across configured voice providers"
+            },
+            "maxAttempts": {
+              "type": "number",
+              "default": 2,
+              "description": "Maximum realtime provider attempts per voice session token request"
+            }
+          }
+        },
         "delegateExecutor": {
           "type": "string",
-          "enum": ["codex-sdk", "copilot-sdk", "claude-sdk", "gemini-sdk", "opencode-sdk"],
+          "enum": [
+            "codex-sdk",
+            "copilot-sdk",
+            "claude-sdk",
+            "gemini-sdk",
+            "opencode-sdk"
+          ],
           "description": "Which executor to use for delegate_to_agent calls. Defaults to primaryAgent."
         }
       }