roadmap.md

APi eeror?

API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}

Not showng Session 213d643d-fc52-4d43-83cd-d4d1e1b45fc6 logs /Users/kirilldubovitskiy/.happy/logs/2025-07-20-20-48-12.log

July 20

3. permission cancelling
4. nice to have - live activity
5. possible crash when disconnects?

Push notification on result Test how permission mode is exited from when in remote

Expired permission requests?

How do we know that Claude is waiting for us in interactive session

On decline - what happens? On timeout?

Cli version report

Not kirill 4) permission ui

updateMetadata - update with usage

• Permission request times out

Failed to get or create session - crashes the cli Ideally we want remote mode to be optional. Should do this with lazy initialization

show we are stuck - likely waiting for permissions?

Edge cases:

• When we are stuck in permissions - we are unable to text & get a response

• User write lol.txt

• Permission blocks for 5 minutes

• If they don't respond within 4.5 minutes,

  • we should abort ourselves, keep the permission up on the client
    • Lets look at async generator to abort
  • When they approve permission next time
  • We should continue

• We should cache permissions previously approved - so on repeat requests - we will auto approve? - later

? Where is the timeout actually coming from?

• Is it claude or mcp server configuration?

• If MCP has defualt timeouts

  • Trying from MCP debug tool - timeout maxx start at 2:35

• Calling interrupt & blocking on mcp server does not yield results. MIght have to do it at the same time

[14:09:13.746] [MessageQueue] waitForNext() adding waiter. Total waiters: 1 [14:13:40.716] [claudeRemote] Received message from SDK: user [14:13:40.717] [CLAUDE] Message from non interactive & remote mode: { "type": "user", "message": { "role": "user", "content": [ { "type": "tool_result", "content": "Error calling tool", "is_error": true, "tool_use_id": "toolu_01BHgTzQcaoa7KMq8sMe1HU1" } ] }, "parent_tool_use_id": null, "session_id": "8226266e-9953-4ea2-b6f1-17384bb4d469" }

Anoterh sample

[14:13:45.662] [CLAUDE] Message from non interactive & remote mode:

14:18:46.683 - error

Basically exactly 5 minutes.

? How do we get the expiration? ? We should reset the permissions?

Or we should allow them to

Now it fails after 5 minutes and tries again with a different permissions? or same

We want to

[15:01:29.481] [SOCKET] Sending message through socket: { "role": "agent", "content": { "type": "output", "data": { "cwd": "/Users/kirilldubovitskiy/projects/happy/handy-cli", "sessionId": "329df624-b37c-4849-ab83-65722d321c29", "version": "1.0.51", "uuid": "072e0a05-5480-4a83-9f63-802da615b66b", "timestamp": "2025-07-20T22:01:29.464Z", "type": "assistant", "message": { "id": "msg_01H9en68JmbGQJb3Mfob2rs8", "type": "message", "role": "assistant", "model": "claude-opus-4-20250514", "content": [ { "type": "text", "text": "I'll create the file lol.txt in the parent directory." } ], "stop_reason": null, "stop_sequence": null }, "requestId": "req_011CRK2MXhzGmZ255atcGVzF" } } }

[15:39:05.950] [SOCKET] Sending message through socket: { "role": "agent", "content": { "type": "output", "data": { "cwd": "/Users/kirilldubovitskiy/projects/happy/handy-cli", "sessionId": "329df624-b37c-4849-ab83-65722d321c29", "version": "1.0.51", "uuid": "8b1593e4-c56e-4785-9dd0-741587021c95", "timestamp": "2025-07-20T22:39:03.317Z", "type": "assistant", "message": { "id": "msg_01H9en68JmbGQJb3Mfob2rs8", "type": "message", "role": "assistant", "model": "claude-opus-4-20250514", "content": [ { "type": "text", "text": "I'll create the file lol.txt in the parent directory." } ], "stop_reason": null, "stop_sequence": null } } } }

• 'result' message type is not sent to history. Its only for sdk

• Remote is maintaned - Starting claudeRemote with messages:

I found the MCP permission tool call timeout configuration!

Key findings:

1. Timeout Configuration Location: The timeout is configured via the RC6() function which reads from the environment variable MCP_TOOL_TIMEOUT: function RC6() { return parseInt(process.env.MCP_TOOL_TIMEOUT || "", 10) || 1e8 }
2. Default Timeout: If MCP_TOOL_TIMEOUT is not set, it defaults to 1e8 milliseconds (100,000,000 ms = ~27.8 hours)
3. Where it's used: This timeout is passed to the MCP client's callTool method in the Uq2 function: let G = await A.callTool({ name: Q, arguments: D }, vm, { signal: I, timeout: RC6() });
4. Connection Timeout: There's also a separate connection timeout configured via Kq2(): function Kq2() { return parseInt(process.env.MCP_TIMEOUT || "", 10) || 30000 }
5. This defaults to 30 seconds and is used for establishing the initial MCP server connection.

To configure the timeout, you can set the environment variable:

• MCP_TOOL_TIMEOUT - for individual tool/permission calls (defaults to ~27.8 hours)

• MCP_TIMEOUT - for initial connection timeout (defaults to 30 seconds)

• interrupt

July 19

• Lazy initialize our shit - immediatelly drop to claude. If we are ofline - still do it

• Passthrough claude parameters

• Server diying test

  • lsof -ti tcp:3005 | xargs kill -9
  • This kills the app :D and cli

• Interruptions add UI to stop (stop button)

• Show that its doing someting, fix thingking

UI

• Permission request
  • Bash
  • Edit / Create (show too much info)
• Tools
  • MultiEdit
  • Task

Nice to have

• Chat titles - big ux boost
• Proxy to show the token count so we know its doing something

Nice nice to have

• Embed amphetamine into it?? -> adderal

July 18

CLI

• Permissions fix

• Test what happens when we timeout the response, how

• Test end to end & rollout new version

CLI dies with error Command failed with exit code 137. info Visit https://yarnpkg.com/en/docs/cli/run for documentation about this command. kirilldubovitskiy@MacBookPro handy-cli % node:events:496 throw er; // Unhandled 'error' event ^

Error: read EIO at TTY.onStreamRead (node:internal/stream_base_commons:216:20) Emitted 'error' event on ReadStream instance at: at emitErrorNT (node:internal/streams/destroy:170:8) at emitErrorCloseNT (node:internal/streams/destroy:129:3) at process.processTicksAndRejections (node:internal/process/task_queues:90:21) { errno: -5, code: 'EIO', syscall: 'read' }

• Dogfood for issues

• Refactor existing code

• Make it feel nicer

• Embed amphetamine into it?

• Permissions - fix mcp server integration

• Deep link to a website

App

• Logout - make a full reload
• Make scroll work nicely

Big ideas

• Coordinator agent - will ensure claude keeps working at max token usage - juice the most out of it
• Social component
• Notifications
• Real time voice

Archive July 18

Roadmap

App

• Make key messages render

• [later] Wrapping claude in an http proxy, allows us to snoop on token usage to show its doing something in the ui when running in remote mode

• For local mode, same approach will work

• Distribution

  • Website - happyinc.ai?
  • App Store
  • Google Play

• Deep link to download app from cli link

Server

• Session management
  • Keep track of who is controlling the session - remote or local

CLI

• Make it stable to be a drop in replacement for claude
• Fix snooping on existing conversation bug, after switching back and forth stops watching the session file for new messages
• [later] Test it works on linux, windows, lower node version

Conversation continuity

• Some things will not expect as you would want such as /clear ing the conversation, or forking (press 2 escape on empty input)
• We might want to be better at switching between sessions for full compatibility with claude

MCP

• Permissions
  • I think we should reuse the format from .claude/settings.local.json, so interactive & our checking will be similar
  • Impelement checking logic
  • Implement blessing command logic ()
• Implement conversation naming
  • I wonder if the server can initiate an llm call on its own accord?

Permission automatic checking

• Pull antropic token from secrets

Blocking

• Permission checking [steve]

  • use mcp
  • see if it has a timeout or we can block forever (ideally)
  • copy cc system (deterministic splitting, prefix checking, injection detection, prefix whitelist suggest)
  • use cc settings local file & format for compatibility
  • figure out extra path permissions

• CLI dies if server disconnects :D

• Need to make agent state work. Most important state - permissions

• Try logging out of Claude and see how to handle that case

• Make sure to use Claude from our package. Kill other Claudes

• Make sure interruption of remote controlled session works

Nice to have

• UX final touches - onboarding make sure terminal, add session icons or something catchy
• See if I can simplify / get rid of a likely race condition in pty related code
• Pass --local-installation to setup .happy folder locally and avoid clashing with global installation

Distribution

• Post on hacker news
• Send to friends to try
• Send to influencers who reviewed similar products
• Mass email people who have starred claudecodeui

Later, low priority

• Permissions callout:

  • permission checking will not be visible on the client nor will we be aware of it
  • ✻ Enchanting… (5s · ↑ 27 tokens · esc to interrupt)
    • We can parse the terminal output

• e2e single tests

  • Would be nice to be able to run the whole thing - including pty to emulate a simple scenario and make sure a single multi step happy path works fine