Merge branch 'dev' into feat/plus-button-sticky

2 files changed, 171 insertions, 53 deletions

diff --git a/packages/core/src/tools/send-to-tab.ts b/packages/core/src/tools/send-to-tab.ts
index eb86b7e..eae6bfa 100644
--- a/packages/core/src/tools/send-to-tab.ts
+++ b/packages/core/src/tools/send-to-tab.ts
@@ -44,6 +44,13 @@ export interface SendToTabCallbacks {
 	/** The calling tab's own id + handle — used to block self-sends and to
 	 *  stamp provenance onto the delivered message. */
 	self: { id: string; handle: string };
+	/**
+	 * Whether THIS calling tab also has the `read_tab` tool granted. The
+	 * tab-messaging permissions are split, so a tab can hold `send_to_tab`
+	 * without `read_tab`. When false, the tool must NOT tell the agent to use
+	 * `read_tab` (it doesn't have it) — replies only arrive on their own.
+	 */
+	canReadTab: boolean;
 }
 
 /** Render the "available tabs" hint shared by the none/ambiguous branches. */
@@ -54,6 +61,19 @@ function renderOpenHandles(handles: Array<{ handle: string; title: string }>): s
 }
 
 export function createSendToTabTool(callbacks: SendToTabCallbacks): ToolDefinition {
+	// The `read_tab` follow-up hint is only truthful when this tab actually
+	// holds the `read_tab` tool (the permissions are split). When it doesn't,
+	// the only honest guidance is that a reply will wake it as a new message — never tell
+	// the agent to call a tool it wasn't granted.
+	const waitLine = callbacks.canReadTab
+		? "money. If the target replies it will WAKE you with a new message in a later turn; you"
+		: "money. If the target replies it will WAKE you with a new message in a later turn.";
+	const readTabLine = callbacks.canReadTab
+		? ["can also call 'read_tab' with the same ID in a FUTURE turn to check. If you have other"]
+		: [];
+	const keepGoingLine = callbacks.canReadTab
+		? "work to do, keep going; if you are ONLY waiting for the reply, end your turn now."
+		: "If you have other work to do, keep going; if you are ONLY waiting for the reply, end your turn now.";
 	return {
 		name: "send_to_tab",
 		description: [
@@ -64,9 +84,14 @@ export function createSendToTabTool(callbacks: SendToTabCallbacks): ToolDefiniti
 			"  - If the target tab is idle, your message WAKES it and starts a new turn.",
 			"",
 			"This is fire-and-forget: it returns immediately and does NOT wait for a reply.",
-			"Use the 'read_tab' tool with the same ID later to read the target's latest response.",
+			"Do NOT sleep, poll, or run shell commands to wait for a reply — that wastes turns and",
+			waitLine,
+			...readTabLine,
+			keepGoingLine,
 			"",
-			"Your tab ID is auto-added to the top of the message so the recipient can reply to you.",
+			"Your tab ID is auto-added to the top of the message so the recipient knows who to reply",
+			"to. The recipient must use this same 'send_to_tab' tool (addressed to your ID) to answer;",
+			"a plain text response reaches only their own user, not you.",
 			"IDs are git-style prefixes: pass any length that uniquely identifies the target (min 4 chars).",
 			"If the ID is ambiguous you'll be asked to add a character.",
 		].join("\n"),
@@ -117,8 +142,18 @@ export function createSendToTabTool(callbacks: SendToTabCallbacks): ToolDefiniti
 			}
 
 			// Stamp provenance so the recipient (and the watching user) can see
-			// which tab the message came from and reply back via its handle.
-			const delivered = `[message from tab ${callbacks.self.handle}]\n\n${message}`;
+			// which tab the message came from and how to reply. The header makes
+			// clear this is a PEER AGENT, not the recipient's own user, and the
+			// footer states the reply contract: a reply (only if warranted) must
+			// go back through `send_to_tab`, since a plain text answer reaches
+			// only the recipient's own user — not this sender.
+			const delivered = [
+				`[message from tab ${callbacks.self.handle} — this is another agent, NOT your user]`,
+				"",
+				message,
+				"",
+				`[To reply to tab ${callbacks.self.handle}, use the send_to_tab tool with tab_id "${callbacks.self.handle}". ONLY reply if this message asks you to, or your user tells you to — it may just be context or instructions. A plain text response goes to your own user, not to this agent.]`,
+			].join("\n");
 
 			try {
 				const result = await callbacks.deliver(target.id, delivered);
@@ -138,7 +173,23 @@ export function createSendToTabTool(callbacks: SendToTabCallbacks): ToolDefiniti
 					result.status === "queued"
 						? "queued (target is busy; it will be picked up next turn)"
 						: "delivered (target was idle; a new turn has started)";
-				return `Message ${verb}. Target tab: ${target.handle} (${target.title}). Use read_tab with "${target.handle}" to read its reply later.`;
+				const tail = callbacks.canReadTab
+					? [
+							"Do NOT sleep, poll, or run commands to wait for a reply. If the target replies it",
+							`will WAKE you with a new message later; you can also call read_tab with "${target.handle}"`,
+							"in a FUTURE turn to check. Keep working if you have other tasks; if you are ONLY",
+							"waiting for this reply, end your turn now.",
+						]
+					: [
+							"Do NOT sleep, poll, or run commands to wait for a reply. If the target replies it",
+							"will WAKE you with a new message later. Keep working if you have other tasks; if",
+							"you are ONLY waiting for this reply, end your turn now.",
+						];
+				return [
+					`Message ${verb}. Target tab: ${target.handle} (${target.title}).`,
+					"",
+					...tail,
+				].join("\n");
 			} catch (err) {
 				return `Error delivering message: ${err instanceof Error ? err.message : String(err)}`;
 			}
diff --git a/packages/core/src/tools/summon.ts b/packages/core/src/tools/summon.ts
index 4820e89..cfee8b8 100644
--- a/packages/core/src/tools/summon.ts
+++ b/packages/core/src/tools/summon.ts
@@ -60,10 +60,13 @@ function renderAgentGroup(label: string, agents: AvailableAgent[]): string[] {
  * the disk locations where they live, injected into the summon tool's
  * description.
  *
- * When `userAgentEnabled` is false only subagents are shown (under the
- * generic "Available agents" heading). When it is true, subagents and
- * user agents are listed as two labelled groups so the LLM understands
- * which slugs require `top_level=true`.
+ * `subagentEnabled` and `userAgentEnabled` independently control which
+ * groups are shown — they mirror the `perm_summon` and `perm_user_agent`
+ * permissions respectively:
+ *   - subagents only → generic "Available agents" heading;
+ *   - user agents only → a single user-agent group (top_level is implied);
+ *   - both → two labelled groups so the LLM understands which slugs
+ *     require `top_level=true`.
  *
  * Returns a compact "no agents defined" notice when nothing is visible.
  */
@@ -72,6 +75,7 @@ function buildAgentsCatalog(
 	userAgents: AvailableAgent[],
 	agentDirs: string[],
 	userAgentEnabled: boolean,
+	subagentEnabled: boolean,
 ): string {
 	const lines: string[] = [];
 	lines.push("");
@@ -80,8 +84,9 @@ function buildAgentsCatalog(
 		lines.push(`  - ${d}`);
 	}
 
+	const visibleSubagents = subagentEnabled ? subagents : [];
 	const visibleUserAgents = userAgentEnabled ? userAgents : [];
-	if (subagents.length === 0 && visibleUserAgents.length === 0) {
+	if (visibleSubagents.length === 0 && visibleUserAgents.length === 0) {
 		lines.push("");
 		lines.push("No agent definitions are currently defined.");
 		return lines.join("\n");
@@ -93,12 +98,26 @@ function buildAgentsCatalog(
 	lines.push("and working directory; the 'tools' parameter is ignored.");
 	lines.push("");
 
+	// User-agent-only mode: list just the user agents. top_level is implied
+	// (it is the only thing this grant can spawn), so the heading omits it.
+	if (!subagentEnabled && userAgentEnabled) {
+		lines.push(
+			...renderAgentGroup(
+				"User agents (spawned as independent top-level tabs):",
+				visibleUserAgents,
+			),
+		);
+		return lines.join("\n");
+	}
+
+	// Subagent-only mode: single generic heading.
 	if (!userAgentEnabled) {
-		lines.push(...renderAgentGroup("Available agents:", subagents));
+		lines.push(...renderAgentGroup("Available agents:", visibleSubagents));
 		return lines.join("\n");
 	}
 
-	const subagentLines = renderAgentGroup("Subagents (spawned as child tabs):", subagents);
+	// Both enabled: two labelled groups.
+	const subagentLines = renderAgentGroup("Subagents (spawned as child tabs):", visibleSubagents);
 	const userAgentLines = renderAgentGroup(
 		"User agents (spawned as independent top-level tabs, requires top_level=true):",
 		visibleUserAgents,
@@ -122,9 +141,14 @@ function buildAgentsCatalog(
  * its description; this is information-only — the runtime resolves
  * slugs through `loadAgent` independently.
  *
- * `userAgentEnabled` controls whether the `top_level` parameter and the
- * user-agent catalog are surfaced to the LLM. It mirrors the
- * `perm_user_agent` permission.
+ * `userAgentEnabled` mirrors the `perm_user_agent` permission and
+ * `subagentEnabled` mirrors the `perm_summon` permission. They are
+ * independent: the tool is registered whenever at least one is granted.
+ *   - subagentEnabled only → spawn ordinary subagents (no `top_level`);
+ *   - userAgentEnabled only → spawn ONLY top-level user agents
+ *     (`top_level` is forced on, the `background` knob is dropped, and
+ *     the catalog lists user agents only);
+ *   - both → full behavior (subagents plus `top_level` user agents).
  */
 export function createSummonTool(
 	_defaultWorkingDirectory: string,
@@ -133,39 +157,29 @@ export function createSummonTool(
 	availableUserAgents: AvailableAgent[] = [],
 	agentDirs: string[] = [],
 	userAgentEnabled = false,
+	subagentEnabled = true,
 ): ToolDefinition {
+	// When only the user-agent permission is granted the tool spawns user
+	// agents exclusively: `top_level` is implied (and forced), subagent
+	// mechanics (background, retrieve, parallel work) are irrelevant.
+	const userAgentOnly = userAgentEnabled && !subagentEnabled;
+
 	const catalog = buildAgentsCatalog(
 		availableSubagents,
 		availableUserAgents,
 		agentDirs,
 		userAgentEnabled,
+		subagentEnabled,
 	);
 	const subagentSlugs = availableSubagents.map((a) => a.slug);
 	const userAgentSlugs = availableUserAgents.map((a) => a.slug);
-	const allSlugs = userAgentEnabled ? [...subagentSlugs, ...userAgentSlugs] : subagentSlugs;
+	const allSlugs = userAgentOnly
+		? userAgentSlugs
+		: userAgentEnabled
+			? [...subagentSlugs, ...userAgentSlugs]
+			: subagentSlugs;
 
-	const description = [
-		"Spawn a new child agent to work on a task independently.",
-		"",
-		"By default, blocks until the child agent finishes and returns the result directly.",
-		"Set background=true to return immediately with an agent_id instead — use retrieve to collect the result later.",
-		"",
-		"The child agent runs in its own tab visible to the user. Use the 'retrieve' tool with the returned agent_id to get the result when needed.",
-		"",
-		"Pattern for parallel work:",
-		"  1. Call summon multiple times with background=true to start several agents",
-		"  2. Do your own work or wait",
-		"  3. Call retrieve for each agent_id to collect results",
-		...(userAgentEnabled
-			? [
-					"",
-					"Set top_level=true to spawn an independent user agent — a first-class",
-					"top-level tab with no parent. User agents are fire-and-forget: you get",
-					"an agent_id back but cannot retrieve their result. top_level requires an",
-					"'agent' definition listed under 'User agents' below.",
-				]
-			: []),
-		"",
+	const toolNamesList = [
 		"The 'tools' parameter controls what the child can do. Available tool names:",
 		"  - read_file: Read file contents",
 		"  - read_file_slice: Read a character-range slice of a single line",
@@ -179,11 +193,50 @@ export function createSummonTool(
 		"  - youtube_transcribe: Fetch YouTube video transcripts",
 		"  - send_to_tab: Send a message to another tab/agent by its ID",
 		"  - read_tab: Read another tab/agent's latest response by its ID",
-		"",
-		"The 'agent' parameter is required — every spawned agent must use a definition.",
-		"Tools default to the agent definition's tools, intersected with your own tools (you can't grant capabilities you don't have).",
-		catalog,
-	].join("\n");
+	];
+
+	const description = userAgentOnly
+		? [
+				"Spawn an independent top-level user agent to work on a task.",
+				"",
+				"User agents are first-class top-level tabs with no parent. They are",
+				"fire-and-forget: you get an agent_id back but cannot retrieve their result.",
+				"The user agent runs in its own tab visible to the user.",
+				"",
+				...toolNamesList,
+				"",
+				"The 'agent' parameter is required — every spawned agent must use a definition.",
+				"Tools default to the agent definition's tools, intersected with your own tools (you can't grant capabilities you don't have).",
+				catalog,
+			].join("\n")
+		: [
+				"Spawn a new child agent to work on a task independently.",
+				"",
+				"By default, blocks until the child agent finishes and returns the result directly.",
+				"Set background=true to return immediately with an agent_id instead — use retrieve to collect the result later.",
+				"",
+				"The child agent runs in its own tab visible to the user. Use the 'retrieve' tool with the returned agent_id to get the result when needed.",
+				"",
+				"Pattern for parallel work:",
+				"  1. Call summon multiple times with background=true to start several agents",
+				"  2. Do your own work or wait",
+				"  3. Call retrieve for each agent_id to collect results",
+				...(userAgentEnabled
+					? [
+							"",
+							"Set top_level=true to spawn an independent user agent — a first-class",
+							"top-level tab with no parent. User agents are fire-and-forget: you get",
+							"an agent_id back but cannot retrieve their result. top_level requires an",
+							"'agent' definition listed under 'User agents' below.",
+						]
+					: []),
+				"",
+				...toolNamesList,
+				"",
+				"The 'agent' parameter is required — every spawned agent must use a definition.",
+				"Tools default to the agent definition's tools, intersected with your own tools (you can't grant capabilities you don't have).",
+				catalog,
+			].join("\n");
 
 	const parametersShape = {
 		task: z
@@ -205,7 +258,10 @@ export function createSummonTool(
 					.filter(Boolean)
 					.join(" "),
 			),
-		...(userAgentEnabled
+		// `top_level` is only an explicit choice when BOTH subagents and user
+		// agents are available. In user-agent-only mode it is implied (forced
+		// on), so the knob is omitted entirely.
+		...(userAgentEnabled && !userAgentOnly
 			? {
 					top_level: z
 						.boolean()
@@ -248,12 +304,18 @@ export function createSummonTool(
 			.describe(
 				"Absolute path for the child to work in. Defaults to the agent definition's cwd (or the spawning agent's directory).",
 			),
-		background: z
-			.boolean()
-			.optional()
-			.describe(
-				"If true, returns immediately with an agent_id for later retrieval. If false (default), blocks until the child agent finishes and returns the result directly. Ignored when top_level is true.",
-			),
+		// `background` is meaningless for fire-and-forget user agents, so the
+		// knob is omitted in user-agent-only mode.
+		...(userAgentOnly
+			? {}
+			: {
+					background: z
+						.boolean()
+						.optional()
+						.describe(
+							"If true, returns immediately with an agent_id for later retrieval. If false (default), blocks until the child agent finishes and returns the result directly. Ignored when top_level is true.",
+						),
+				}),
 	};
 
 	return {
@@ -266,9 +328,14 @@ export function createSummonTool(
 			const tools = args.tools as string[] | undefined;
 			const workingDirectory = args.working_directory as string | undefined;
 			const background = (args.background as boolean | undefined) ?? false;
-			const topLevel = userAgentEnabled
-				? ((args.top_level as boolean | undefined) ?? false)
-				: false;
+			// User-agent-only mode always spawns top-level user agents. When both
+			// capabilities are present the caller chooses via `top_level`. When
+			// only subagents are available, top-level spawning is unavailable.
+			const topLevel = userAgentOnly
+				? true
+				: userAgentEnabled
+					? ((args.top_level as boolean | undefined) ?? false)
+					: false;
 
 			try {
 				const agentId = await callbacks.spawn({