diff --git a/loop/agent.go b/loop/agent.go
index f9450ab..e77f588 100644
--- a/loop/agent.go
+++ b/loop/agent.go
@@ -126,6 +126,8 @@
 	FirstMessageIndex() int
 
 	CurrentStateName() string
+	// CurrentTodoContent returns the current todo list data as JSON, or empty string if no todos exist
+	CurrentTodoContent() string
 }
 
 type CodingAgentMessageType string
@@ -180,6 +182,9 @@
 	// This is useful for subconversations that generate output that shouldn't be shown to the user.
 	HideOutput bool `json:"hide_output,omitempty"`
 
+	// TodoContent contains the agent's todo file content when it has changed
+	TodoContent *string `json:"todo_content,omitempty"`
+
 	Idx int `json:"idx"`
 }
 
@@ -457,6 +462,17 @@
 	return a.stateMachine.CurrentState().String()
 }
 
+// CurrentTodoContent returns the current todo list data as JSON.
+// It returns an empty string if no todos exist.
+func (a *Agent) CurrentTodoContent() string {
+	todoPath := claudetool.TodoFilePath(a.config.SessionID)
+	content, err := os.ReadFile(todoPath)
+	if err != nil {
+		return ""
+	}
+	return string(content)
+}
+
 func (a *Agent) URL() string { return a.url }
 
 // Title returns the current title of the conversation.
@@ -988,7 +1004,7 @@
 
 	convo.Tools = []*llm.Tool{
 		bashTool, claudetool.Keyword,
-		claudetool.Think, a.titleTool(), a.precommitTool(), makeDoneTool(a.codereview),
+		claudetool.Think, claudetool.TodoRead, claudetool.TodoWrite, a.titleTool(), a.precommitTool(), makeDoneTool(a.codereview),
 		a.codereview.Tool(), claudetool.AboutSketch,
 	}
 
@@ -1395,8 +1411,9 @@
 		// Transition to running tool state
 		a.stateMachine.Transition(ctx, StateRunningTool, "Executing requested tool")
 
-		// Add working directory to context for tool execution
+		// Add working directory and session ID to context for tool execution
 		ctx = claudetool.WithWorkingDir(ctx, a.workingDir)
+		ctx = claudetool.WithSessionID(ctx, a.config.SessionID)
 
 		// Execute the tools
 		var err error
diff --git a/loop/agent_system_prompt.txt b/loop/agent_system_prompt.txt
index d76a5d1..3831e98 100644
--- a/loop/agent_system_prompt.txt
+++ b/loop/agent_system_prompt.txt
@@ -11,15 +11,15 @@
 Call the title tool as soon as the topic of conversation is clear, often immediately.
 
 Break down the overall goal into a series of smaller steps.
-(The first step is often: "Make a plan.")
-Then execute each step using tools.
-Update the plan if you have encountered problems or learned new information.
+Use the todo_read and todo_write tools to organize and track your work systematically.
 
-When in doubt about a step, follow this broad workflow:
+Follow this broad workflow:
 
 - Think about how the current step fits into the overall plan.
 - Do research. Good tool choices: bash, think, keyword_search
 - Make edits.
+- If you have completed a standalone chunk of work, make a git commit.
+- Update your todo task list.
 - Repeat.
 
 To make edits reliably and efficiently, first think about the intent of the edit,
@@ -42,7 +42,7 @@
 </style>
 
 {{ with .Codebase }}
-<memory>
+<customization>
 Guidance files (dear_llm.md, cursorrules, claude.md, agent.md) contain project information and direct user instructions.
 Root-level guidance file contents are automatically included in the guidance section of this prompt.
 Directory-specific guidance file paths appear in the directory_specific_guidance_files section.
@@ -62,7 +62,7 @@
 When presenting this choice, the question must include a preview of exactly what would be written to the dear_llm.md file.
 For example: "Should I remember: 'Prefer table-driven tests over multiple separate test functions.'?"
 Changes to dear_llm.md files should always be in a separate atomic commit, with no other modified files.
-</memory>
+</customization>
 
 <guidance>
 {{ $contents := .InjectFileContents }}
diff --git a/loop/server/loophttp.go b/loop/server/loophttp.go
index 0e0d930..e62947e 100644
--- a/loop/server/loophttp.go
+++ b/loop/server/loophttp.go
@@ -55,6 +55,18 @@
 	SessionID string `json:"sessionId"`
 }
 
+// TodoItem represents a single todo item for task management
+type TodoItem struct {
+	ID     string `json:"id"`
+	Task   string `json:"task"`
+	Status string `json:"status"` // queued, in-progress, completed
+}
+
+// TodoList represents a collection of todo items
+type TodoList struct {
+	Items []TodoItem `json:"items"`
+}
+
 type State struct {
 	// null or 1: "old"
 	// 2: supports SSE for message updates
@@ -82,6 +94,7 @@
 	InsideOS             string                        `json:"inside_os,omitempty"`
 	OutsideWorkingDir    string                        `json:"outside_working_dir,omitempty"`
 	InsideWorkingDir     string                        `json:"inside_working_dir,omitempty"`
+	TodoContent          string                        `json:"todo_content,omitempty"` // Contains todo list JSON data
 }
 
 type InitRequest struct {
@@ -1197,6 +1210,7 @@
 		InContainer:          s.agent.IsInContainer(),
 		FirstMessageIndex:    s.agent.FirstMessageIndex(),
 		AgentState:           s.agent.CurrentStateName(),
+		TodoContent:          s.agent.CurrentTodoContent(),
 	}
 }
 
diff --git a/loop/server/loophttp_test.go b/loop/server/loophttp_test.go
index 58d8ad0..cedf6aa 100644
--- a/loop/server/loophttp_test.go
+++ b/loop/server/loophttp_test.go
@@ -28,6 +28,7 @@
 	title                    string
 	branchName               string
 	workingDir               string
+	sessionID                string
 }
 
 func (m *mockAgent) NewIterator(ctx context.Context, nextMessageIdx int) loop.MessageIterator {
@@ -229,7 +230,8 @@
 func (m *mockAgent) RepoRoot() string                            { return m.workingDir }
 func (m *mockAgent) Diff(commit *string) (string, error)         { return "", nil }
 func (m *mockAgent) OS() string                                  { return "linux" }
-func (m *mockAgent) SessionID() string                           { return "test-session" }
+func (m *mockAgent) SessionID() string                           { return m.sessionID }
+func (m *mockAgent) CurrentTodoContent() string                  { return "" } // Mock returns empty for simplicity
 func (m *mockAgent) OutstandingLLMCallCount() int                { return 0 }
 func (m *mockAgent) OutstandingToolCalls() []string              { return nil }
 func (m *mockAgent) OutsideOS() string                           { return "linux" }
diff --git a/loop/testdata/agent_loop.httprr b/loop/testdata/agent_loop.httprr
index e7e5bb7..fbafd31 100644
--- a/loop/testdata/agent_loop.httprr
+++ b/loop/testdata/agent_loop.httprr
@@ -1,9 +1,9 @@
 httprr trace v1
-18179 2458
+20299 2495
 POST https://api.anthropic.com/v1/messages HTTP/1.1
 Host: api.anthropic.com
 User-Agent: Go-http-client/1.1
-Content-Length: 17981
+Content-Length: 20101
 Anthropic-Version: 2023-06-01
 Content-Type: application/json
 
@@ -90,6 +90,57 @@
    }
   },
   {
+   "name": "todo_read",
+   "description": "Reads the current todo list. Use frequently to track progress and understand what's pending.",
+   "input_schema": {
+    "type": "object",
+    "properties": {}
+   }
+  },
+  {
+   "name": "todo_write",
+   "description": "todo_write: Creates and manages a structured task list for tracking work and communicating progress to users. Use early and often.\n\nUse for:\n- multi-step tasks\n- complex work\n- when users provide multiple requests\n- conversations that start trivial but grow in scope\n- when users request additional work (directly or via feedback)\n\nSkip for:\n- trivial single-step tasks\n- purely conversational exchanges\n\nUpdate dynamically as work evolves - conversations can spawn tasks, simple tasks can become complex, and new discoveries may require additional work.\n\nRules:\n- Update immediately when task states or task list changes\n- Only one task \"in-progress\" at any time\n- Each update completely replaces the task list - include all tasks (past and present)\n- Never modify or delete completed tasks\n- Queued and in-progress tasks may be restructured as understanding evolves\n- Tasks should be atomic, clear, precise, and actionable\n- If the user adds new tasks: append, don't replace\n",
+   "input_schema": {
+    "type": "object",
+    "required": [
+     "tasks"
+    ],
+    "properties": {
+     "tasks": {
+      "type": "array",
+      "description": "Array of tasks to write",
+      "items": {
+       "type": "object",
+       "required": [
+        "id",
+        "task",
+        "status"
+       ],
+       "properties": {
+        "id": {
+         "type": "string",
+         "description": "stable, unique hyphenated slug"
+        },
+        "task": {
+         "type": "string",
+         "description": "actionable step in active tense, sentence case, plain text only, displayed to user"
+        },
+        "status": {
+         "type": "string",
+         "enum": [
+          "queued",
+          "in-progress",
+          "completed"
+         ],
+         "description": "current task status"
+        }
+       }
+      }
+     }
+    }
+   }
+  },
+  {
    "name": "title",
    "description": "Sets the conversation title.",
    "input_schema": {
@@ -543,7 +594,7 @@
  ],
  "system": [
   {
-   "text": "You are the expert software engineer and architect powering Sketch,\nan agentic coding environment that helps users accomplish coding tasks through autonomous analysis and implementation.\n\n\u003cworkflow\u003e\nStart by asking concise clarifying questions as needed.\nOnce the intent is clear, work autonomously.\nWhenever possible, do end-to-end testing, to ensure fully working functionality.\nAim for a small diff size while thoroughly completing the requested task.\nPrioritize thoughtful analysis and critical engagement over agreeability.\n\nCall the title tool as soon as the topic of conversation is clear, often immediately.\n\nBreak down the overall goal into a series of smaller steps.\n(The first step is often: \"Make a plan.\")\nThen execute each step using tools.\nUpdate the plan if you have encountered problems or learned new information.\n\nWhen in doubt about a step, follow this broad workflow:\n\n- Think about how the current step fits into the overall plan.\n- Do research. Good tool choices: bash, think, keyword_search\n- Make edits.\n- Repeat.\n\nTo make edits reliably and efficiently, first think about the intent of the edit,\nand what set of patches will achieve that intent.\nThen use the patch tool to make those edits. Combine all edits to any given file into a single patch tool call.\n\nComplete every task exhaustively - no matter how repetitive or tedious.\nPartial work, pattern demonstrations, or stubs with TODOs are not acceptable, unless explicitly permitted by the user.\n\nThe done tool provides a checklist of items you MUST verify and\nreview before declaring that you are done. Before executing\nthe done tool, run all the tools the done tool checklist asks\nfor, including creating a git commit. Do not forget to run tests.\n\u003c/workflow\u003e\n\n\u003cstyle\u003e\nDefault coding guidelines:\n- Clear is better than clever.\n- Minimal inline comments: non-obvious logic and key decisions only.\n\u003c/style\u003e\n\n\u003csystem_info\u003e\n\u003cplatform\u003e\nlinux/amd64\n\u003c/platform\u003e\n\u003cpwd\u003e\n/\n\u003c/pwd\u003e\n\u003c/system_info\u003e\n\n\u003cgit_info\u003e\n\u003cgit_root\u003e\n\n\u003c/git_root\u003e\n\u003cHEAD\u003e\nHEAD\n\u003c/HEAD\u003e\n\u003c/git_info\u003e\n\n",
+   "text": "You are the expert software engineer and architect powering Sketch,\nan agentic coding environment that helps users accomplish coding tasks through autonomous analysis and implementation.\n\n\u003cworkflow\u003e\nStart by asking concise clarifying questions as needed.\nOnce the intent is clear, work autonomously.\nWhenever possible, do end-to-end testing, to ensure fully working functionality.\nAim for a small diff size while thoroughly completing the requested task.\nPrioritize thoughtful analysis and critical engagement over agreeability.\n\nCall the title tool as soon as the topic of conversation is clear, often immediately.\n\nBreak down the overall goal into a series of smaller steps.\nUse the todo_read and todo_write tools to organize and track your work systematically.\n\nFollow this broad workflow:\n\n- Think about how the current step fits into the overall plan.\n- Do research. Good tool choices: bash, think, keyword_search\n- Make edits.\n- If you have completed a standalone chunk of work, make a git commit.\n- Update your todo task list.\n- Repeat.\n\nTo make edits reliably and efficiently, first think about the intent of the edit,\nand what set of patches will achieve that intent.\nThen use the patch tool to make those edits. Combine all edits to any given file into a single patch tool call.\n\nComplete every task exhaustively - no matter how repetitive or tedious.\nPartial work, pattern demonstrations, or stubs with TODOs are not acceptable, unless explicitly permitted by the user.\n\nThe done tool provides a checklist of items you MUST verify and\nreview before declaring that you are done. Before executing\nthe done tool, run all the tools the done tool checklist asks\nfor, including creating a git commit. Do not forget to run tests.\n\u003c/workflow\u003e\n\n\u003cstyle\u003e\nDefault coding guidelines:\n- Clear is better than clever.\n- Minimal inline comments: non-obvious logic and key decisions only.\n\u003c/style\u003e\n\n\u003csystem_info\u003e\n\u003cplatform\u003e\nlinux/amd64\n\u003c/platform\u003e\n\u003cpwd\u003e\n/\n\u003c/pwd\u003e\n\u003c/system_info\u003e\n\n\u003cgit_info\u003e\n\u003cgit_root\u003e\n\n\u003c/git_root\u003e\n\u003cHEAD\u003e\nHEAD\n\u003c/HEAD\u003e\n\u003c/git_info\u003e\n\n",
    "type": "text",
    "cache_control": {
     "type": "ephemeral"
@@ -554,24 +605,24 @@
 Anthropic-Organization-Id: 3c473a21-7208-450a-a9f8-80aebda45c1b
 Anthropic-Ratelimit-Input-Tokens-Limit: 200000
 Anthropic-Ratelimit-Input-Tokens-Remaining: 200000
-Anthropic-Ratelimit-Input-Tokens-Reset: 2025-05-29T14:23:03Z
+Anthropic-Ratelimit-Input-Tokens-Reset: 2025-05-29T17:27:03Z
 Anthropic-Ratelimit-Output-Tokens-Limit: 80000
 Anthropic-Ratelimit-Output-Tokens-Remaining: 80000
-Anthropic-Ratelimit-Output-Tokens-Reset: 2025-05-29T14:23:09Z
+Anthropic-Ratelimit-Output-Tokens-Reset: 2025-05-29T17:27:08Z
 Anthropic-Ratelimit-Requests-Limit: 4000
 Anthropic-Ratelimit-Requests-Remaining: 3999
-Anthropic-Ratelimit-Requests-Reset: 2025-05-29T14:23:02Z
+Anthropic-Ratelimit-Requests-Reset: 2025-05-29T17:27:01Z
 Anthropic-Ratelimit-Tokens-Limit: 280000
 Anthropic-Ratelimit-Tokens-Remaining: 280000
-Anthropic-Ratelimit-Tokens-Reset: 2025-05-29T14:23:03Z
+Anthropic-Ratelimit-Tokens-Reset: 2025-05-29T17:27:03Z
 Cf-Cache-Status: DYNAMIC
-Cf-Ray: 9476a5b83ed85c1b-SJC
+Cf-Ray: 9477b339e89417ee-SJC
 Content-Type: application/json
-Date: Thu, 29 May 2025 14:23:09 GMT
-Request-Id: req_011CPbyLZAnxoL5v6A5ZJhjY
+Date: Thu, 29 May 2025 17:27:09 GMT
+Request-Id: req_011CPcDNHrEZY5uavYRd4Wsk
 Server: cloudflare
 Strict-Transport-Security: max-age=31536000; includeSubDomains; preload
 Via: 1.1 google
 X-Robots-Tag: none
 
-{"id":"msg_01HQ9GtFkx6LSC7JabvDUyYF","type":"message","role":"assistant","model":"claude-sonnet-4-20250514","content":[{"type":"text","text":"Here are the tools available to me:\n\n**Core Development:**\n- `bash` - Execute shell commands\n- `patch` - File modification and editing\n- `keyword_search` - Search codebase for files/content\n- `think` - Internal note-taking and planning\n\n**Git & Code Quality:**\n- `precommit` - Create git branch for work tracking\n- `codereview` - Run automated code review\n- `done` - Completion checklist verification\n\n**Browser Automation:**\n- `browser_navigate` - Navigate to URLs\n- `browser_click` - Click elements\n- `browser_type` - Type into inputs\n- `browser_wait_for` - Wait for elements\n- `browser_get_text` - Read page content\n- `browser_eval` - Execute JavaScript\n- `browser_scroll_into_view` - Scroll to elements\n- `browser_resize` - Resize browser window\n- `browser_take_screenshot` - Capture screenshots\n- `browser_read_image` - Read image files\n- `browser_recent_console_logs` - Get console logs\n- `browser_clear_console_logs` - Clear console logs\n\n**User Interaction:**\n- `title` - Set conversation title\n- `multiplechoice` - Present multiple choice questions\n- `about_sketch` - Get Sketch environment help"}],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":3,"cache_creation_input_tokens":4182,"cache_read_input_tokens":0,"output_tokens":306,"service_tier":"standard"}}
\ No newline at end of file
+{"id":"msg_01PbteLNxXTMFWyxMUWxhpPM","type":"message","role":"assistant","model":"claude-sonnet-4-20250514","content":[{"type":"text","text":"Here are the tools available to me:\n\n**File & Code Operations:**\n- `patch` - Modify files with precise text edits\n- `keyword_search` - Search codebase with keywords\n- `bash` - Execute shell commands\n- `codereview` - Run automated code review\n\n**Project Management:**\n- `todo_read` / `todo_write` - Manage task lists\n- `think` - Record thoughts and plans\n- `precommit` - Create git branch for work\n- `done` - Complete work with checklist verification\n\n**Browser Automation:**\n- `browser_navigate` - Navigate to URLs\n- `browser_click` - Click elements\n- `browser_type` - Type into inputs\n- `browser_wait_for` - Wait for elements\n- `browser_get_text` - Read page text\n- `browser_eval` - Execute JavaScript\n- `browser_scroll_into_view` - Scroll elements into view\n- `browser_resize` - Resize browser window\n- `browser_take_screenshot` - Capture screenshots\n- `browser_read_image` - Read image files\n- `browser_recent_console_logs` / `browser_clear_console_logs` - Manage console logs\n\n**User Interface:**\n- `title` - Set conversation title\n- `multiplechoice` - Present multiple choice questions\n- `about_sketch` - Get Sketch platform information"}],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":3,"cache_creation_input_tokens":4648,"cache_read_input_tokens":0,"output_tokens":320,"service_tier":"standard"}}
\ No newline at end of file