Reproduced on Linux web dev at 8acbefefc5, using Claude Opus 5.5 in Supervised mode. Requested a Bash command writing a temporary marker with the description “Write the triage marker”; the approval detail showed only that description, without the command. Existing #20145 addresses this path, so this triage preserves that work. Desktop/Electron remains unverified.
[Bug]: Aproval prompts do not show the command being approved#20146
Before submitting
- I searched existing issues and did not find a duplicate.
- I included enough detail to reproduce or investigate the problem.
Area
apps/desktop
Steps to reproduce
- Start a chat with supervised (or auto-accept edits) permissions with a prompt that requires running tools/commands that need manual approval.
- Wait until the agent attempts to run a command and you're asked for permission.
Expected behavior
I can easily see enough details to confidently decide if the action is safe, preferably including the exact action or command that is about to be taken.
Actual behavior
I get Claude Code's description which (I think) is generated by the model calling the command. This is too vague for an informed decision and might be entirely incorrect.
Impact
Major degradation or frequent failure
Version or commit
0.1.0
Environment
Desktop App with Claude Code provider running locally
Logs or stack traces
shell
| 1 |
Screenshots, recordings, or supporting files
No response
Workaround
I only run threads on machines where I can use auto permissions so that I don't have to manually approve anything.