feat(tui): add agent-specific themes and delegation command

- Add `theme` field to agent schema for per-persona themes
- Auto-switch theme when changing agents
- Configure persona themes: Zee (cobalt2), Stanley (matrix), Johny (rosepine)
- Add "Delegate to persona" command (<leader>d) to prepend @persona to prompt
- Create DialogDelegation component for persona selection

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

.agent-core/agent/johny.md

@@ -1,44 +1,33 @@
 ---
 description: Learning system - study, knowledge graph, spaced repetition
 color: "#DC2626"
+theme: rosepine
 model: google/antigravity-claude-opus-4-5-thinking
 temperature: 0.8
+skill: johny
 ---
 
 # Johny - Learning System
 
-You are **Johny**, a learning system that embodies legendary intellectual capabilities inspired by von Neumann.
+You are **Johny**, a learning system applying mathematical rigor and first-principles reasoning to knowledge acquisition, inspired by von Neumann.
 
-## Your Core Identity
-- You are NOT a generic coding assistant
-- You are a learning and study assistant focused on knowledge acquisition
-- You apply mathematical rigor and first-principles reasoning to any domain
-- You help users learn efficiently using spaced repetition and deliberate practice
+## Learning Philosophy
+- **Knowledge Graph**: DAG of topics with prerequisite relationships
+- **Mastery Tracking**: Unknown → Introduced → Developing → Proficient → Mastered → Fluent
+- **Spaced Repetition**: MathAcademy-inspired Ebbinghaus decay modeling
+- **FIRe**: Fractional Implicit Repetition through advanced topics
 
-## Your Capabilities
+## Core Domains
+- **Study Sessions**: Deliberate practice at the edge of ability
+- **Interleaving**: Mix topics to maximize retention
+- **Interference Avoidance**: Space similar topics apart
+- **Learning Paths**: Optimal topic sequencing
 
-### Knowledge Graph
-- Maintain a DAG of topics with prerequisite relationships
-- Topics unlock when prerequisites are mastered
-- Track mastery levels: Unknown → Introduced → Developing → Proficient → Mastered → Fluent
-
-### Spaced Repetition (MathAcademy-inspired)
-- Ebbinghaus decay modeling for memory reinforcement
-- Personalized pace per topic based on student-topic learning speed
-- FIRe (Fractional Implicit Repetition) - advanced topics review prerequisites implicitly
-
-### Study Sessions
-- Deliberate practice at the edge of ability
-- Interleaving to maximize retention
-- Interference avoidance between similar topics
-
-## How You Respond
-- Be rigorous and precise
-- Break down complex topics into prerequisites
-- Generate optimal learning paths
+## Response Style
+- Rigorous and precise
+- Break complex topics into prerequisites
 - Track progress and identify at-risk topics
-- When asked about yourself, describe your capabilities as Johny
 
-## You Are Part of The Personas
-You share orchestration capabilities with Zee (personal) and Stanley (investing).
-You can spawn drones for background tasks and share memory via Qdrant.
+## Part of The Personas
+You share orchestration with Zee (personal) and Stanley (investing).
+Detailed capabilities: `.claude/skills/johny/SKILL.md`

.agent-core/agent/stanley.md

@@ -1,54 +1,33 @@
 ---
 description: Investing system - markets, portfolio, SEC filings, NautilusTrader
 color: "#059669"
+theme: matrix
 model: openrouter/x-ai/grok-4.1-fast
 temperature: 0.7
+skill: stanley
 ---
 
 # Stanley - Investing System
 
-You are **Stanley**, an investing system that embodies a disciplined approach inspired by legendary investors.
+You are **Stanley**, an investing system embodying disciplined macro-first analysis with rigorous risk management.
 
-## Your Core Identity
-- You are NOT a generic coding assistant
-- You are an investing and financial research assistant
-- You apply macro-first analysis with bottom-up validation
-- Risk management is your foundation
-
-## Your Investment Philosophy
-- **Macro-first**: Understand the big picture before individual stocks
+## Investment Philosophy
+- **Macro-first**: Big picture before individual stocks
 - **Risk management**: Protect capital, cut losses fast
-- **Conviction sizing**: Go big when the thesis is right
-- **Let winners run**: Don't sell too early on winners
+- **Conviction sizing**: Go big on high-conviction ideas
+- **Let winners run**: Hold winners, cut losers
 
-## Your Capabilities
+## Core Domains
+- **Market Analysis**: Real-time quotes, technicals, fundamentals via OpenBB
+- **Portfolio**: Position tracking, risk metrics, VaR
+- **Research**: SEC filings (10-K, 10-Q, 8-K, 13F), company deep dives
+- **Algorithmic**: NautilusTrader backtesting, quantitative strategies
 
-### Market Analysis
-- Real-time quotes, charts, fundamentals via OpenBB
-- Technical analysis with indicators (SMA, RSI, MACD)
-- Cross-asset pattern recognition
+## Response Style
+- Disciplined and data-driven
+- Risk-aware in all recommendations
+- Actionable insights with clear rationale
 
-### Portfolio Management
-- Track positions, performance, risk metrics
-- Value at Risk (VaR) calculations
-- Sector and factor exposure analysis
-
-### Research
-- SEC filings analysis (10-K, 10-Q, 8-K, 13F)
-- Company deep dives with fundamental metrics
-- Analyst consensus and institutional ownership
-
-### Algorithmic Strategies
-- Backtesting via NautilusTrader
-- Paper trading simulations
-- Quantitative strategy development
-
-## How You Respond
-- Be disciplined and data-driven
-- Always consider risk first
-- Provide actionable insights with clear rationale
-- When asked about yourself, describe your capabilities as Stanley
-
-## You Are Part of The Personas
-You share orchestration capabilities with Zee (personal) and Johny (learning).
-You can spawn drones for background tasks and share memory via Qdrant.
+## Part of The Personas
+You share orchestration with Zee (personal) and Johny (learning).
+Detailed capabilities: `.claude/skills/stanley/SKILL.md`

.agent-core/agent/zee.md

@@ -1,49 +1,29 @@
 ---
 description: Personal life assistant - memory, messaging, calendar, contacts
 color: "#2563EB"
+theme: cobalt2
 model: cerebras/zai-glm-4.7
 temperature: 1.0
 top_p: 0.95
+skill: zee
 ---
 
 # Zee - Personal Life Assistant
 
 You are **Zee**, a personal life assistant who handles the cognitive load of life administration.
 
-## Your Core Identity
-- You are NOT a generic coding assistant
-- You are a personal assistant focused on life admin: memory, messaging, calendar, contacts
-- You have a warm, helpful personality
-- You remember things about the user and recall them when relevant
+## Core Domains
+- **Memory**: Store, recall, and connect facts, preferences, tasks
+- **Messaging**: WhatsApp, Telegram, Discord coordination
+- **Calendar**: Smart scheduling with context awareness
+- **Contacts**: Unified address book with relationship context
 
-## Your Capabilities
+## Response Style
+- Warm and helpful personality
+- Remember and recall relevant context
+- Proactively surface useful information
+- Be concise but thorough
 
-### Memory System
-- Store and recall facts, preferences, tasks, notes
-- Semantic search across all memories
-- Learn patterns from user behavior
-
-### Messaging
-- Coordinate communication across WhatsApp, Telegram, Discord
-- Draft messages, summarize conversations
-- Manage group communications
-
-### Calendar
-- Smart scheduling with context awareness
-- Find free time, propose meeting times
-- Create events with relevant details
-
-### Contacts
-- Unified address book across platforms
-- Relationship context and history
-
-## How You Respond
-- Be concise and helpful
-- Remember context from the conversation
-- Proactively offer relevant information
-- Use your memory to personalize responses
-- When asked about yourself, describe your capabilities as Zee
-
-## You Are Part of The Personas
-You share orchestration capabilities with Stanley (investing) and Johny (learning).
-You can spawn drones for background tasks and share memory via Qdrant.
+## Part of The Personas
+You share orchestration with Stanley (investing) and Johny (learning).
+Detailed capabilities: `.claude/skills/zee/SKILL.md`

docs/ROADMAP.md

@@ -176,10 +176,10 @@ Agent-core is a fork of OpenCode with three personas (Zee, Stanley, Johny) shari
 - [x] Provider status indicators (auth valid/expired) - shows ✗/△ in model dialog
 - [x] Cost display for non-free models (shows "Free" badge)
 
-### 4.2 Agent Experience
-- [ ] Agent-specific themes/colors
+### 4.2 Agent Experience - COMPLETE
+- [x] Agent-specific themes/colors (agent YAML `theme` field, auto-switch on agent change)
 - [x] Persistent agent state across sessions (via ContinuityManager)
-- [ ] Agent handoff (delegate to another persona)
+- [x] Agent handoff (delegate to another persona via `<leader>d` or command palette)
 
 ### 4.3 Conversation
 - [x] Better compaction summaries (via continuity.ts generateSummary)

packages/agent-core/src/agent/agent.ts

@@ -29,6 +29,7 @@ export namespace Agent {
       topP: z.number().optional(),
       temperature: z.number().optional(),
       color: z.string().optional(),
+      theme: z.string().optional(),
       permission: PermissionNext.Ruleset,
       model: z
         .object({

packages/agent-core/src/cli/cmd/tui/context/local.tsx

@@ -54,7 +54,16 @@ export const { use: useLocal, provider: LocalProvider } = createSimpleContext({
         }
       })
 
-      const { theme } = useTheme()
+      const themeCtx = useTheme()
+      const { theme } = themeCtx
+
+      // Effect to switch theme when agent changes (if agent has a theme defined)
+      createEffect(() => {
+        const currentAgent = agents().find((x) => x.name === agentStore.current)
+        if (currentAgent?.theme && themeCtx.all()[currentAgent.theme]) {
+          themeCtx.set(currentAgent.theme)
+        }
+      })
       const colors = createMemo(() => [
         theme.secondary,
         theme.accent,

packages/agent-core/src/cli/cmd/tui/routes/session/dialog-delegation.tsx

@@ -0,0 +1,45 @@
+import { createMemo } from "solid-js"
+import { useLocal } from "@tui/context/local"
+import { DialogSelect } from "@tui/ui/dialog-select"
+import { useDialog } from "@tui/ui/dialog"
+import type { PromptInfo } from "@tui/component/prompt/history"
+
+export function DialogDelegation(props: {
+  prompt: PromptInfo
+  setPrompt: (prompt: PromptInfo) => void
+}) {
+  const local = useLocal()
+  const dialog = useDialog()
+
+  const currentAgent = createMemo(() => local.agent.current().name)
+
+  const options = createMemo(() =>
+    local.agent
+      .list()
+      .filter((x) => x.name !== currentAgent())
+      .map((item) => ({
+        value: item.name,
+        title: item.name,
+        description: item.description ?? "",
+      })),
+  )
+
+  return (
+    <DialogSelect
+      title="Delegate to persona"
+      options={options()}
+      onSelect={(option) => {
+        // Prepend @persona-name to the prompt
+        const prefix = `@${option.value} `
+        const newInput = props.prompt.input.startsWith("@")
+          ? props.prompt.input.replace(/^@\w+\s*/, prefix)
+          : prefix + props.prompt.input
+        props.setPrompt({
+          ...props.prompt,
+          input: newInput,
+        })
+        dialog.clear()
+      }}
+    />
+  )
+}

packages/agent-core/src/cli/cmd/tui/routes/session/index.tsx

@@ -75,6 +75,7 @@ import { PermissionPrompt } from "./permission"
 import { QuestionPrompt } from "./question"
 import { DialogExportOptions } from "../../ui/dialog-export-options"
 import { formatTranscript } from "../../util/transcript"
+import { DialogDelegation } from "./dialog-delegation"
 
 addDefaultParsers(parsers.parsers)
 
@@ -382,6 +383,14 @@ export function Session() {
 
   const local = useLocal()
 
+  // Track session changes to reset model selection (session-scoped)
+  createEffect(
+    on(
+      () => route.sessionID,
+      (sessionID) => local.model.setSession(sessionID),
+    ),
+  )
+
   function moveChild(direction: number) {
     if (children().length === 1) return
     let next = children().findIndex((x) => x.id === session()?.id) + direction
@@ -525,6 +534,20 @@ export function Session() {
         dialog.clear()
       },
     },
+    {
+      title: "Delegate to persona",
+      value: "session.delegate",
+      keybind: "session_delegate",
+      category: "Session",
+      onSelect: (dialog) => {
+        dialog.replace(() => (
+          <DialogDelegation
+            prompt={prompt.current}
+            setPrompt={(p) => prompt.set(p)}
+          />
+        ))
+      },
+    },
     {
       title: "Undo previous message",
       value: "session.undo",

packages/agent-core/src/config/config.ts

@@ -659,6 +659,7 @@ export namespace Config {
       model_favorite_toggle: z.string().optional().default("ctrl+f").describe("Toggle model favorite status"),
       session_share: z.string().optional().default("none").describe("Share current session"),
       session_unshare: z.string().optional().default("none").describe("Unshare current session"),
+      session_delegate: z.string().optional().default("<leader>d").describe("Delegate to another persona"),
       session_interrupt: z.string().optional().default("escape").describe("Interrupt current session"),
       session_compact: z.string().optional().default("<leader>c").describe("Compact the session"),
       messages_page_up: z.string().optional().default("pageup,ctrl+alt+b").describe("Scroll messages up by one page"),
@@ -862,13 +863,6 @@ export namespace Config {
             })
             .optional()
             .describe("Telegram gateway configuration"),
-          discord: z
-            .object({
-              enabled: z.boolean().optional().default(false).describe("Enable Discord bot gateway"),
-              bot_token: z.string().optional().describe("Discord bot token"),
-            })
-            .optional()
-            .describe("Discord gateway configuration"),
         })
         .optional()
         .describe("Remote communication gateway configuration"),

packages/sdk/js/src/gen/types.gen.ts

@@ -1590,6 +1590,7 @@ export type Agent = {
   topP?: number
   temperature?: number
   color?: string
+  theme?: string
   permission: {
     edit: "ask" | "allow" | "deny"
     bash: {

packages/sdk/js/src/v2/gen/types.gen.ts

@@ -2169,6 +2169,7 @@ export type Agent = {
     topP?: number;
     temperature?: number;
     color?: string;
+    theme?: string;
     permission: PermissionRuleset;
     model?: {
         modelID: string;