Add investigation item for context event vs before_agent_start

Reference: #324

Documents:
- Current context event behavior (AgentMessage[], transient)
- Proposed before_agent_start event (persistent, TUI visible)
- Key differences table
- Open design questions
- Need to verify AgentMessage vs Message abstraction level

packages/ai/src/models.generated.ts

@@ -6104,9 +6104,9 @@ export const MODELS = {
 			contextWindow: 32768,
 			maxTokens: 4096,
 		} satisfies Model<"openai-completions">,
-		"anthropic/claude-3.5-haiku-20241022": {
+		"anthropic/claude-3.5-haiku": {
-			id: "anthropic/claude-3.5-haiku-20241022",
+			id: "anthropic/claude-3.5-haiku",
-			name: "Anthropic: Claude 3.5 Haiku (2024-10-22)",
+			name: "Anthropic: Claude 3.5 Haiku",
 			api: "openai-completions",
 			provider: "openrouter",
 			baseUrl: "https://openrouter.ai/api/v1",
@@ -6121,9 +6121,9 @@ export const MODELS = {
 			contextWindow: 200000,
 			maxTokens: 8192,
 		} satisfies Model<"openai-completions">,
-		"anthropic/claude-3.5-haiku": {
+		"anthropic/claude-3.5-haiku-20241022": {
-			id: "anthropic/claude-3.5-haiku",
+			id: "anthropic/claude-3.5-haiku-20241022",
-			name: "Anthropic: Claude 3.5 Haiku",
+			name: "Anthropic: Claude 3.5 Haiku (2024-10-22)",
 			api: "openai-completions",
 			provider: "openrouter",
 			baseUrl: "https://openrouter.ai/api/v1",
@@ -6359,23 +6359,6 @@ export const MODELS = {
 			contextWindow: 128000,
 			maxTokens: 16384,
 		} satisfies Model<"openai-completions">,
-		"meta-llama/llama-3.1-8b-instruct": {
-			id: "meta-llama/llama-3.1-8b-instruct",
-			name: "Meta: Llama 3.1 8B Instruct",
-			api: "openai-completions",
-			provider: "openrouter",
-			baseUrl: "https://openrouter.ai/api/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 0.02,
-				output: 0.03,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 131072,
-			maxTokens: 16384,
-		} satisfies Model<"openai-completions">,
 		"meta-llama/llama-3.1-405b-instruct": {
 			id: "meta-llama/llama-3.1-405b-instruct",
 			name: "Meta: Llama 3.1 405B Instruct",
@@ -6410,6 +6393,23 @@ export const MODELS = {
 			contextWindow: 131072,
 			maxTokens: 4096,
 		} satisfies Model<"openai-completions">,
+		"meta-llama/llama-3.1-8b-instruct": {
+			id: "meta-llama/llama-3.1-8b-instruct",
+			name: "Meta: Llama 3.1 8B Instruct",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 0.02,
+				output: 0.03,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 131072,
+			maxTokens: 16384,
+		} satisfies Model<"openai-completions">,
 		"mistralai/mistral-nemo": {
 			id: "mistralai/mistral-nemo",
 			name: "Mistral: Mistral Nemo",
@@ -6546,23 +6546,6 @@ export const MODELS = {
 			contextWindow: 128000,
 			maxTokens: 4096,
 		} satisfies Model<"openai-completions">,
-		"openai/gpt-4o-2024-05-13": {
-			id: "openai/gpt-4o-2024-05-13",
-			name: "OpenAI: GPT-4o (2024-05-13)",
-			api: "openai-completions",
-			provider: "openrouter",
-			baseUrl: "https://openrouter.ai/api/v1",
-			reasoning: false,
-			input: ["text", "image"],
-			cost: {
-				input: 5,
-				output: 15,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 128000,
-			maxTokens: 4096,
-		} satisfies Model<"openai-completions">,
 		"openai/gpt-4o": {
 			id: "openai/gpt-4o",
 			name: "OpenAI: GPT-4o",
@@ -6597,6 +6580,23 @@ export const MODELS = {
 			contextWindow: 128000,
 			maxTokens: 64000,
 		} satisfies Model<"openai-completions">,
+		"openai/gpt-4o-2024-05-13": {
+			id: "openai/gpt-4o-2024-05-13",
+			name: "OpenAI: GPT-4o (2024-05-13)",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			reasoning: false,
+			input: ["text", "image"],
+			cost: {
+				input: 5,
+				output: 15,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 128000,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
 		"meta-llama/llama-3-70b-instruct": {
 			id: "meta-llama/llama-3-70b-instruct",
 			name: "Meta: Llama 3 70B Instruct",
@@ -6835,23 +6835,6 @@ export const MODELS = {
 			contextWindow: 8191,
 			maxTokens: 4096,
 		} satisfies Model<"openai-completions">,
-		"openai/gpt-4": {
-			id: "openai/gpt-4",
-			name: "OpenAI: GPT-4",
-			api: "openai-completions",
-			provider: "openrouter",
-			baseUrl: "https://openrouter.ai/api/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 30,
-				output: 60,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 8191,
-			maxTokens: 4096,
-		} satisfies Model<"openai-completions">,
 		"openai/gpt-3.5-turbo": {
 			id: "openai/gpt-3.5-turbo",
 			name: "OpenAI: GPT-3.5 Turbo",
@@ -6869,6 +6852,23 @@ export const MODELS = {
 			contextWindow: 16385,
 			maxTokens: 4096,
 		} satisfies Model<"openai-completions">,
+		"openai/gpt-4": {
+			id: "openai/gpt-4",
+			name: "OpenAI: GPT-4",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			reasoning: false,
+			input: ["text"],
+			cost: {
+				input: 30,
+				output: 60,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 8191,
+			maxTokens: 4096,
+		} satisfies Model<"openai-completions">,
 		"openrouter/auto": {
 			id: "openrouter/auto",
 			name: "OpenRouter: Auto Router",

packages/coding-agent/docs/session-tree-plan.md

@@ -285,6 +285,59 @@ Benefits:
 - Works with branching (pruning entries are part of the tree)
 - Trade-off: cache busting on first submission after pruning
 
+### Investigate: `context` event vs `before_agent_start`
+
+Reference: [#324](https://github.com/badlogic/pi-mono/issues/324)
+
+**Current `context` event:**
+- Fires before each LLM call within the agent loop
+- Receives `AgentMessage[]` (deep copy, safe to modify)
+- Modifications are transient (not persisted to session)
+- No TUI visibility of what was changed
+- Use case: non-destructive pruning, dynamic context manipulation
+
+**Problem:** `AgentMessage` includes custom types (hookMessage, bashExecution, etc.) that need conversion to LLM `Message[]` before sending. Need to verify:
+- [ ] Where does `AgentMessage[]` → `Message[]` conversion happen relative to `context` event?
+- [ ] Should hooks work with `AgentMessage[]` or `Message[]`?
+- [ ] Is the current abstraction level correct?
+
+**Proposed `before_agent_start` event:**
+- Fires once when user submits a prompt, before `agent_start`
+- Allows hooks to inject additional content that gets **persisted** to session
+- Injected content is visible in TUI (observability)
+- Does not bust prompt cache (appended after user message, not modifying system prompt)
+
+**Key difference:**
+| Aspect | `context` | `before_agent_start` |
+|--------|-----------|---------------------|
+| When | Before each LLM call | Once per user prompt |
+| Persisted | No | Yes (as SystemMessage) |
+| TUI visible | No | Yes (collapsible) |
+| Cache impact | Can bust cache | Append-only, cache-safe |
+| Use case | Transient manipulation | Persistent context injection |
+
+**Design questions:**
+- [ ] Should `before_agent_start` create a new message type (`SystemMessage` with `role: "system"`)?
+- [ ] How should it render in TUI? (label when collapsed, full content when expanded)
+- [ ] How does it interact with compaction? (treated like user messages?)
+- [ ] Can hook return multiple messages or just one?
+
+**Implementation sketch:**
+```typescript
+interface BeforeAgentStartEvent {
+  type: "before_agent_start";
+  userMessage: UserMessage;  // The prompt user just submitted
+}
+
+interface BeforeAgentStartResult {
+  /** Additional context to inject (persisted as SystemMessage) */
+  inject?: {
+    label: string;           // Shown in collapsed TUI state
+    content: string | (TextContent | ImageContent)[];
+  };
+}
+```
+
 ### HTML Export
 
 - [ ] Add collapsible sidebar showing full tree structure