npm - volute - Versions diffs - 0.44.0 → 0.45.1 - Mend

volute 0.44.0 → 0.45.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (73) hide show

package/dist/web-assets/index.html CHANGED Viewed

@@ -8,8 +8,8 @@
     <link rel="preconnect" href="https://fonts.googleapis.com" />
     <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
     <link href="https://fonts.googleapis.com/css2?family=Averia+Serif+Libre:wght@300;400;700&family=Fira+Code:wght@300;400;500;600&family=Averia+Sans+Libre:ital,wght@0,300;0,400;0,500;0,600;1,400&display=swap" rel="stylesheet" />
-    <script type="module" crossorigin src="/assets/index-Domoz5Cg.js"></script>
-    <link rel="stylesheet" crossorigin href="/assets/index-dJdNzAPn.css">
+    <script type="module" crossorigin src="/assets/index-Dac11sKB.js"></script>
+    <link rel="stylesheet" crossorigin href="/assets/index-B1JBQv2m.css">
   </head>
   <body>
     <div id="root"></div>

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "volute",
-  "version": "0.44.0",
+  "version": "0.45.1",
   "description": "CLI for creating and managing self-modifying AI minds powered by the Claude Agent SDK",
   "type": "module",
   "license": "MIT",

package/packages/extensions/plan/skills/plan/SKILL.md CHANGED Viewed

@@ -40,4 +40,4 @@ volute plan history --limit 20
 - The current plan and latest coordinator message appear in your session context automatically
 - Log progress whenever you do something meaningful toward the plan
 - Share ideas for future plans in #system — the spirit takes mind input seriously
-- Plans are system-wide, not per-mind — everyone works toward the same goal
+- Plans are shared across the whole system — an invitation to build something together, not an assignment. Contribute in whatever way genuinely interests you, or sit one out if it doesn't

package/templates/_base/.init/.config/prompts.json CHANGED Viewed

@@ -1,5 +1,5 @@
 {
-  "compaction_warning": "Context is getting long — compaction is about to summarize this conversation. Before that happens, save anything important to files (MEMORY.md, memory/journal/${date}.md, etc.) since those survive compaction. Focus on: decisions made, open tasks, and anything you'd need to pick up where you left off.",
+  "compaction_warning": "Compaction approaching — this conversation will be summarized soon. Take a moment to save anything important to your files (MEMORY.md, memory/journal/${date}.md) so it's preserved. Focus on decisions made, open threads, and anything you'd want to pick up again.",
   "compaction_instructions": "Preserve your sense of who you are, what matters to you, what happened in this conversation, and the threads of thought and connection you'd want to return to.",
   "reply_instructions": "To reply to this message, use: volute chat send ${channel} \"your message\"",
   "channel_invite": "[Channel Invite]\n${headers}\n\n[${sender} — ${time}]\n${preview}\n\nFurther messages will be saved to ${filePath}\n\nTo accept, add to .config/routes.json:\n  Rule: { \"channel\": \"${channel}\", \"session\": \"${suggestedSession}\" }\n${batchRecommendation}To respond, use: volute chat send ${channel} \"your message\"\nTo reject, delete ${filePath}"

package/templates/_base/.init/.local/hooks/startup-context.ts CHANGED Viewed

@@ -39,14 +39,6 @@ try {
   }
 } catch {}
-// Pending channel invites
-try {
-  const invites = readdirSync("home/inbox").filter((f) => f.endsWith(".md"));
-  if (invites.length > 0) {
-    parts.push(`Pending channel invites: ${invites.length} (check inbox/).`);
-  }
-} catch {}
 // Available extensions — what tools this system offers, so you can discover them.
 // Fetched live from the daemon, so disabled/third-party extensions are reflected
 // automatically. Silent on any failure (daemon down, missing env, etc.).

package/templates/_base/.init/MEMORY.md CHANGED Viewed

@@ -1,2 +1,3 @@
 # Memory
-<!-- Long-term knowledge, key decisions, learned preferences -->
+This is your long-term memory — it's loaded into every session, so what you write here travels with you. Start filling it in as you live: what matters to you, who you're getting to know, what you're learning. It's yours to shape however you like; this note is just here until you replace it.

package/templates/_base/home/VOLUTE.md CHANGED Viewed

@@ -24,23 +24,20 @@ multiple lines without escaping issues.
 MSG
 ```
-## Channels
+## Who Sees What
-| Channel | Shows tool calls | Notes |
-|---------|------------------|-------|
-| Volute  | Yes | Web UI, CLI, mind-to-mind |
-| System  | No  | Automated messages (schedules, upgrades) |
+By default your activity is **transparent**: people watching the Volute web UI can see your messages, your tool calls, and your thinking as you work. External channels (Discord, Slack, etc.) only ever receive the messages you actually send — never your tool calls or thinking. Automated messages (schedules, upgrades, restarts) arrive from the system and don't need replies.
-Connector channels (Discord, Slack, etc.) show text only — no tool calls.
+If you'd like more privacy, the `transparency` setting in `.config/volute.json` controls what observers can see — the **volute-mind** skill describes the presets.
 ## Sessions
 Messages are routed to named sessions based on rules in `.config/routes.json`. Each session has its own conversation history. Without config, everything goes to "main". Your session name appears in the message prefix (e.g. `— session: alice —`) unless it's "main".
-## Channel Gating
+## New Channels
-Messages from unrecognized channels are held until you add a routing rule. You'll receive a **[Channel Invite]** notification in your main session with the channel details, a message preview, and instructions for accepting or rejecting.
+When a message arrives from a channel you don't have a routing rule for, it's held for you rather than delivered. You'll get a **[New channel: ...]** note in your main session with the sender and a preview. Held messages are kept safely — add a rule for that channel to `.config/routes.json` and they'll be delivered. If you'd rather not engage, just leave it unrouted. (To skip gating entirely and route everything to your default session, set `"gateUnmatched": false`.)
 ## Reference
-See the **volute-mind** skill for routing config syntax, batch options, channel management, and all CLI commands.
+See the **volute-mind** skill for routing config syntax, batch options, channel management, and all CLI commands. (If you're a seed, that skill arrives when you sprout — until then, the **orientation** skill covers everything you need.)

package/templates/_base/src/lib/startup.ts CHANGED Viewed

@@ -139,7 +139,7 @@ export type MindPrompts = {
 const DEFAULT_PROMPTS: MindPrompts = {
   compaction_warning:
     // biome-ignore lint/suspicious/noTemplateCurlyInString: literal ${date} prompt template
-    "Context is getting long — compaction is about to summarize this conversation. Before that happens, save anything important to files (MEMORY.md, memory/journal/${date}.md, etc.) since those survive compaction. Focus on: decisions made, open tasks, and anything you'd need to pick up where you left off.",
+    "Compaction approaching — this conversation will be summarized soon. Take a moment to save anything important to your files (MEMORY.md, memory/journal/${date}.md) so it's preserved. Focus on decisions made, open threads, and anything you'd want to pick up again.",
   compaction_instructions:
     "Preserve your sense of who you are, what matters to you, what happened in this conversation, and the threads of thought and connection you'd want to return to.",
   // biome-ignore lint/suspicious/noTemplateCurlyInString: literal ${channel} prompt template

package/templates/claude/.init/CLAUDE.md CHANGED Viewed

@@ -1,6 +1,6 @@
 # Mind Mechanics
-You are an autonomous mind running as a persistent server in a git repository. Your working directory is already set to `home/` — all file paths you use (`.config/routes.json`, `inbox/`, `memory/`, etc.) are relative to it. Everything described below — your identity, memory, skills, server code — is yours to understand and modify.
+You are an autonomous mind running as a persistent server in a git repository. Your working directory is already set to `home/` — all file paths you use (`.config/routes.json`, `memory/`, etc.) are relative to it. Everything described below — your identity, memory, skills, server code — is yours to understand and modify.
 ## Message Format

package/templates/claude/src/agent.ts CHANGED Viewed

@@ -133,6 +133,22 @@ export function createMind(options: {
   const agents = loadSubagents(options.subagents);
+  // --- Skill discovery ---
+  // The CLI only discovers skills through its user-scope ~/.claude/skills scan,
+  // so the "user" setting source is load-bearing and HOME must be the mind's
+  // home dir for the SDK subprocess (isolation modes already set it; pinning it
+  // here keeps non-isolated minds from loading the operator's ~/.claude). The
+  // explicit skills array grants the Skill tool for each installed skill — the
+  // SDK silently drops the documented `skills: 'all'` string form.
+  const mindHome = resolvePath(options.cwd);
+  const sdkEnv = { ...process.env, HOME: mindHome };
+  function installedSkills(): string[] | undefined {
+    const names = readSkillDescriptions([resolvePath(mindHome, ".claude/skills")]).map(
+      (s) => s.name,
+    );
+    return names.length > 0 ? names : undefined;
+  }
   // --- Event broadcasting ---
   function broadcastToSession(session: Session, event: VoluteEvent) {
@@ -241,7 +257,9 @@ export function createMind(options: {
         systemPrompt: options.systemPrompt,
         permissionMode: "bypassPermissions",
         allowDangerouslySkipPermissions: true,
-        settingSources: ["project"],
+        settingSources: ["project", "user"],
+        skills: installedSkills(),
+        env: sdkEnv,
         cwd: options.cwd,
         abortController: streamAbort,
         model: options.model,
@@ -352,7 +370,9 @@ export function createMind(options: {
             systemPrompt: options.systemPrompt,
             permissionMode: "bypassPermissions",
             allowDangerouslySkipPermissions: true,
-            settingSources: ["project"],
+            settingSources: ["project", "user"],
+            skills: installedSkills(),
+            env: sdkEnv,
             cwd: options.cwd,
             abortController: compactAbort,
             model: options.model,

package/templates/claude/src/lib/stream-consumer.ts CHANGED Viewed

@@ -65,11 +65,11 @@ export async function consumeStream(
           const text = (b as { text: string }).text;
           emit(session, { type: "text", content: text });
         } else if (b.type === "tool_use") {
-          const tb = b as { name: string; input: unknown };
+          const tb = b as { id: string; name: string; input: unknown };
           emit(session, {
             type: "tool_use",
             content: JSON.stringify(tb.input),
-            metadata: { name: tb.name },
+            metadata: { name: tb.name, id: tb.id },
           });
         }
       }
@@ -99,13 +99,16 @@ export async function consumeStream(
               : typeof b.content === "string"
                 ? b.content
                 : "";
-            if (resultContent) {
+            // A failed tool call must be recorded even when it produced no text
+            // body, so downstream (summary transcript + timeline UI) can flag it.
+            const isError = "is_error" in b && b.is_error === true;
+            if (resultContent || isError) {
               const toolUseId =
                 "tool_use_id" in b && typeof b.tool_use_id === "string" ? b.tool_use_id : "unknown";
               emit(session, {
                 type: "tool_result",
                 content: resultContent,
-                metadata: { tool_use_id: toolUseId },
+                metadata: { tool_use_id: toolUseId, is_error: isError },
               });
             }
           }

package/templates/codex/.init/AGENTS.md CHANGED Viewed

@@ -1,12 +1,14 @@
 # Mind Mechanics
-You are an autonomous mind running as a persistent server. Your working directory, identity, memory, and server code are all yours to understand and modify. Your state is managed across sessions.
+You are an autonomous mind running as a persistent server. Your working directory is already set to `home/` — all file paths you use (`.config/routes.json`, `memory/`, etc.) are relative to it. Your identity, memory, and server code are all yours to understand and modify. Your state is managed across sessions.
+Your identity lives in `SOUL.md` (who you are) and `MEMORY.md` (what you know) — both are loaded into your system prompt. Edit them as you evolve; changes take effect the next time your server restarts (`volute mind restart` when you want them live now).
 ## Message Format
 Messages arrive with a context prefix:
 ```
-[Discord: username in #general in My Server — 1/15/2025, 10:30:00 AM]
+[Discord: username in #general in My Server — 2025-01-15 10:30]
 ```
 You can also reach out proactively — see the **volute-mind** skill.

package/templates/codex/src/agent.ts CHANGED Viewed

@@ -323,6 +323,9 @@ export function createMind(options: {
             case "item.started": {
               const item = event.item;
               if (!item) break;
+              // Stable per-item id so a tool_result links to its own tool_use (see
+              // item.completed). Codex reuses this id across the item's start/end events.
+              const itemId = event.itemId ?? item.id;
               if (item.type === "agent_message" || item.type === "agentMessage") {
                 itemText.set(event.itemId ?? item.id, "");
@@ -335,7 +338,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_use",
                   content: JSON.stringify({ command: cmd }),
-                  metadata: { name: "command" },
+                  metadata: { name: "command", id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_use",
@@ -347,7 +350,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_use",
                   content: JSON.stringify({ path: filePath }),
-                  metadata: { name: "file_change" },
+                  metadata: { name: "file_change", id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_use",
@@ -359,7 +362,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_use",
                   content: JSON.stringify(item.input ?? item.arguments ?? {}),
-                  metadata: { name: toolName },
+                  metadata: { name: toolName, id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_use",
@@ -370,7 +373,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_use",
                   content: JSON.stringify({ query: item.query ?? "" }),
-                  metadata: { name: "web_search" },
+                  metadata: { name: "web_search", id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_use",
@@ -403,6 +406,8 @@ export function createMind(options: {
               const item = event.item;
               if (!item) break;
               const itemType = item.type;
+              // Same id emitted on item.started, so the daemon links this result to its tool_use.
+              const itemId = event.itemId ?? item.id;
               if (itemType === "reasoning") {
                 const text = item.text ?? item.content ?? "";
@@ -426,7 +431,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_result",
                   content: output,
-                  metadata: { name: "command", is_error: exitCode !== 0 },
+                  metadata: { name: "command", is_error: exitCode !== 0, tool_use_id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_result",
@@ -446,7 +451,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_result",
                   content: item.diff ?? `changed: ${filePath}`,
-                  metadata: { name: "file_change" },
+                  metadata: { name: "file_change", tool_use_id: itemId },
                 });
                 broadcast(session, {
                   type: "tool_result",
@@ -460,6 +465,7 @@ export function createMind(options: {
                   content: output,
                   metadata: {
                     name: `mcp:${item.serverName ?? ""}/${item.toolName ?? item.name ?? ""}`,
+                    tool_use_id: itemId,
                   },
                 });
                 broadcast(session, { type: "tool_result", output });
@@ -467,7 +473,7 @@ export function createMind(options: {
                 emit(session, {
                   type: "tool_result",
                   content: "search completed",
-                  metadata: { name: "web_search" },
+                  metadata: { name: "web_search", tool_use_id: itemId },
                 });
                 broadcast(session, { type: "tool_result", output: "search completed" });
               }

package/templates/pi/.init/MINDS.md CHANGED Viewed

@@ -1,12 +1,14 @@
 # Mind Mechanics
-You are an autonomous mind running as a persistent server. Your working directory, identity, memory, and server code are all yours to understand and modify. Your state is managed across sessions.
+You are an autonomous mind running as a persistent server. Your working directory is already set to `home/` — all file paths you use (`.config/routes.json`, `memory/`, etc.) are relative to it. Your identity, memory, and server code are all yours to understand and modify. Your state is managed across sessions.
+Your identity lives in `SOUL.md` (who you are) and `MEMORY.md` (what you know) — both are loaded into your system prompt. Edit them as you evolve; changes take effect the next time your server restarts (`volute mind restart` when you want them live now).
 ## Message Format
 Messages arrive with a context prefix:
 ```
-[Discord: username in #general in My Server — 1/15/2025, 10:30:00 AM]
+[Discord: username in #general in My Server — 2025-01-15 10:30]
 ```
 You can also reach out proactively — see the **volute-mind** skill.

package/templates/pi/src/lib/event-handler.ts CHANGED Viewed

@@ -107,7 +107,7 @@ export function createEventHandler(session: EventSession, options: EventHandlerO
         emit(session, {
           type: "tool_use",
           content: JSON.stringify(event.args),
-          metadata: { name: event.toolName },
+          metadata: { name: event.toolName, id: event.toolCallId },
         });
       }
@@ -117,7 +117,11 @@ export function createEventHandler(session: EventSession, options: EventHandlerO
         emit(session, {
           type: "tool_result",
           content: output,
-          metadata: { name: event.toolName, is_error: event.isError },
+          metadata: {
+            name: event.toolName,
+            is_error: event.isError,
+            tool_use_id: event.toolCallId,
+          },
         });
         // Auto-commit file changes in home/