Claude Auth Triage

[lukehinds]

Claude Auth Triage

Only use this if Claude Code is running inside nono, is currently logged out, and this same instance was previously logged in and should have refreshed automatically but instead is incorrectly asking you to run /login again.

If your setup is currently working, do not run this.

Linux users should run this too. The capture includes some macOS-specific checks, but they are skipped or ignored automatically on Linux.

Before changing config, reinstalling tools, or trying workarounds, capture the failure once first.

Run This

Run this only if the auth failure is happening right now or is immediately reproducible.

command -v nono
nono --version
curl -fsSL https://gist.githubusercontent.com/lukehinds/275325a85d5649f18cbc3a187e5a52d5/raw/3b47192da8a9766d5281d148243a63c2657de6f9/capture-live-claude-auth-failure.sh -o /tmp/capture-live-claude-auth-failure.sh
chmod +x /tmp/capture-live-claude-auth-failure.sh
/tmp/capture-live-claude-auth-failure.sh \
  --label discord-user-auth-failure \
  --nono-bin "$(command -v nono)"

Claude may open twice during the capture:

• first for a startup probe
• then for the real repro capture

Each time Claude opens:

• do not run /login
• do not try multiple different things in the same run
• run /exit

Then attach the archive file produced by the script to this GitHub discussion thread as a comment

If command -v nono prints nothing, stop there and report that first.

macOS Version Note

If nono --version reports 0.34.0, 0.35.0, 0.36.0, or 0.37.0, mention that in your reply.

Those versions contained a real macOS keychain Seatbelt regression for Claude auth.

claude-code installation method

Please include details of how you installed claude, and if you're a linux user, the current dist or package manager used.

[a3lem]

Here you go!

❯ nono --version
nono 0.37.0

20260420-124106-discord-user-auth-failure.tar.gz

[romgX]

This looks like a session state mismatch issue 🔍

I've hit this exact scenario before. The problem is usually that Claude Code's cached auth token exists locally but the nono environment doesn't recognize it as valid anymore—so instead of silently refreshing, it keeps asking you to re-authenticate.

A few things to check before running that capture script:

1. Token location: Where is your auth token stored? Check:

   ls -la ~/.claude/  # or wherever nono expects it
   stat ~/.claude/auth.json  # check if it's being written to

2. Environment variables: Make sure NONO_HOME or similar isn't pointing somewhere unexpected:

   env | grep -i nono
   env | grep -i claude

3. Quick reset (if you want to try before the full diagnostic):

   rm ~/.claude/auth.json  # or the actual token path
   # Then open Claude Code again and do /login once cleanly

That capture script is solid for detailed debugging—it'll show you exactly what's happening during the auth flow. Just make sure:

• Claude isn't actively trying to authenticate when you run it
• You let it open twice without interrupting
• Save the output so you can see the actual error

What OS are you on? Linux users sometimes hit different token path issues than macOS, so that'll help narrow down next steps.

[rnestler]

Why the --label discord-user-auth-failure? Probably a copy paste from a different debugging session?

[lukehinds]

Its from when I had a thread in discord and asked folks to capture , but it does look a bit weird without that context. Are you having login issues?

[rnestler]

I get an error with the script on my system:

[2026-05-05 09:42:39] Capturing current environment state
script: unexpected number of arguments
Try 'script --help' for more information.

The issue seems to be that it tries to start script like this:

script -q /tmp/nono-live-claude-auth-failure/20260505-094453-claude-auth-failure/env-capture/startup-probe.txt /tmp/nono-live-auth-cmd.btGSm5

But (at least on my system) script wants a -- to separate the command:

script -q /tmp/nono-live-claude-auth-failure/20260505-094453-claude-auth-failure/env-capture/startup-probe.txt -- /tmp/nono-live-auth-cmd.btGSm5

[rnestler]

Its from when I had a thread in discord and asked folks to capture , but it does look a bit weird without that context. Are you having login issues?

Yes. When running under nono claude asks a lot for login: Please run /login · API Error: 401 {"type":"error","error":{"type":"authentication_error","message":"Invalid authentication credentials"},"request_id":"req_xxx"}

But it may also be related to my claude-code installation method. I use the AUR package. This also meant I needed to create a separate profile for it to allow it to start its own binary:

{
  "extends": "claude-code",
  "meta": {
    "name": "claude"
  },
  "security": {
    "groups": []
  },
  "workdir": {
    "access": "readwrite"
  },
  "filesystem": {
    "allow": ["/tmp/claude-1000/", "/tmp/claude-gather/"],
    "read": ["/opt/claude-code/"]
  }
}

/usr/bin/claude is just a simple wrapper script:

/usr/bin/claude
#!/bin/sh
export DISABLE_UPDATES=1
exec /opt/claude-code/bin/claude "$@"

[rnestler]

After editing the script I managed to run it:

20260505-110725-claude-auth-failure.tar.gz