Troubleshooting

Copilot says "No authentication information found"

This is the most common report, and it is a Copilot session-expiry issue rather than a Frontier bug. Copilot's non-interactive mode only works while the CLI is currently logged in. Login state lives in ~/.copilot/config.json under loggedInUsers. If that array is empty — with only lastLoggedInUser present — the CLI is logged out and every headless run fails.

copilot login

A provider shows Ready but tasks fail

Ready only means <exe> --version succeeded — the binary was found. It says nothing about whether the CLI has a valid session. Run the CLI yourself in a terminal; if it prompts you to log in, that is your answer.

A provider shows Not detected

Frontier could not locate the executable on PATH. Find the absolute path and paste it into that provider's Executable field:

command -v codex
command -v claude
command -v copilot

This is a PATH or executable-configuration issue. It is not a filesystem permission issue — a packaged app does not inherit your shell's PATH the way a terminal does.

A task failed and did not fail over

That is intended. Only quota, rate-limit, overload, and unavailable errors move work to another agent. A genuine agent failure stops the task, because re-running a partially completed coding task through a different agent could duplicate or conflict with edits already written to disk. Cancellation is terminal too. See routing & failover.

Codex rejects the model I picked

Model ids are CLI-specific. Codex fails an entire run when handed a Claude model id such as claude-opus-5 ("not supported when using Codex with a ChatGPT account"). Frontier tags the model you pick with the agent it came from and only ever hands it to that agent — every other provider runs its own model and notes it in the transcript. If you see this error, the model was configured directly on the Codex provider card.

Usage looks wrong or missing

• Claude shows a window but no percentage. Correct — Claude reports no utilisation figure. The muted bar tracks elapsed time in the window, and is labelled as such.
• Copilot and Ollama show nothing. Correct — they stream no JSON to report from.
• A window vanished. Windows past their reset time are dropped rather than held at a stale percentage.

The synthesis step redid all the work

Subtask changes are committed to their own frontier/* branches, so they are not in your working tree when the final report is written. Frontier instructs the synthesizer that this is a read-only reporting step for exactly that reason. Check the Review screen — the work is on the branches.

macOS will not open the app

Builds are unsigned. Open it the first time with Control-click → Open and confirm. On Windows, choose More info → Run anyway in SmartScreen.

Packaging fails on Windows

pnpm dist can fail through its pnpm build && … prefix when pnpm re-runs install. Build the steps directly instead:

npm run build
./node_modules/.bin/electron-builder

If you hit EPERM: … rename 'release\win-unpacked', close any running Frontier Proxy instance, delete release/win-unpacked*, and re-run.

Still stuck?

Open an issue with your operating system, the Frontier version from the bottom of the sidebar, which agent failed, and the CLI's own output — github.com/ThisaST/Frontier-Proxy/issues.