Desktop app ignores `askUser: false` from settings.json and offers no way to disable the ask_user tool
- Dominant language
- Shell
- Stars
- 11.2k
- Forks
- 1.9k
- Avg merge
- 14h 16m
- Merged PRs (30d)
- 6
Description
> *This was generated by AI during triage.*
## Summary
The desktop app has no way to disable the `ask_user` tool. The CLI setting `askUser: false` in `~/.copilot/settings.json` is read by the CLI entry point only; the app is a separate host that never consults it and exposes no equivalent toggle, so the same configuration directory yields different agent behavior depending on which client opens the session.
## Environment
- GitHub Copilot desktop app v1.0.26
- App-spawned Copilot CLI runtime v1.0.71
- Copilot CLI v1.0.75
- Windows 11
- `~/.copilot/settings.json` contains `"askUser": false`
## Reproduction
1. Set `"askUser": false` in `~/.copilot/settings.json`.
2. Start `copilot` in a terminal and confirm the agent has no `ask_user` tool.
3. Open the desktop app against the same configuration directory and start any chat or project session.
4. Observe that the agent still has `ask_user` and uses it.
## Expected behavior
One of the following, in order of preference:
- The app honors `askUser` from `~/.copilot/settings.json`, matching the CLI.
- The app exposes its own setting that maps to the SDK session option `askUserDisabled` (or `defaultAgent.excludedTools: ["ask_user"]`, which the app's own session wire already carries).
- The app documents that `askUser` is CLI-only, so the setting is not silently inert.
## Sanitized evidence
Read-only inspection of the installed binaries:
```text
CLI 1.0.75 app.js
askUserDisabled: (…) || settings.askUser === false # prompt mode
askUserDisabled: !B # interactive mode, B from --ask-user/--no-ask-user
option("--no-ask-user", "Disable the ask_user tool (agent works autonomously without asking questions)")
Desktop app 1.0.26 github.exe
SessionCreateWire fields: availableTools, excludedTools, defaultAgent, … # no askUserDisabled
string "askUser" — no occurrences
chat session tool set (Rust, src-tauri/src/session/core.rs region):
ask_user glob rg skill view web_fetch web_search
```
- `~/.copilot/data.db` tables `settings` and `app_state` hold no tool-permission or `askUser` state.
- The app does read `~/.copilot/config.json` and `~/.copilot/settings.json` (for example `skillDirectories`), so the file is in scope for the app; only this key is unhandled.
## Observed versus unconfirmed
**Observed:** `askUser: false` has no effect in the app, and no app setting, app-state key, or database column controls it.
**Unconfirmed:** whether a custom agent whose `tools:` allowlist omits `ask_user` fully suppresses the tool for a main app session — the SDK types describe `CustomAgentConfig.tools` as an allowlist, but this was not exercised end to end. Custom instructions can only discourage the tool, not remove it.
## Why it matters
Autonomy is a per-user working preference, not a per-client one. Users who deliberately run a no-questions agent lose that guarantee the moment the work moves from the terminal to the app, and the two clients disagree while reading the same configuration directory.
## Questions
1. Is `askUser` intended to remain CLI-only, or should host-independent settings apply to every client that reads `~/.copilot`?
2. Would `defaultAgent.excludedTools` be the preferred app-side mechanism, given the session wire already supports it?
3. Should settings the app cannot honor be surfaced as inert rather than ignored silently?
## Additional context
This was a read-only investigation: binaries, the configuration directory, and a copy of `data.db` were inspected; nothing was modified.
Contributor guide
Research direction
Start in src-tauri/src/session/core.rs and inspect how the desktop session builds its tool set and SessionCreateWire fields. Compare that path with the CLI handling of askUser in the inspected app.js evidence. Done means the app consistently honors askUser: false or provides and documents an equivalent setting, with the resulting session verified to exclude ask_user.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- cli, desktop, tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100