Troubleshooting
Start with atlas dev --status (what is running and in what state) and atlas version. Logs for every service are in <project>/.logs/.
Install and update#
| Symptom | Cause and fix |
|---|---|
atlas: command not found after installing | Open a new shell: atlas install edits your shell’s startup file. Check ~/.atlas/bin is on PATH. |
macOS kills the binary (zsh: killed) | The binary is not signed. Release builds are ad-hoc signed; for a binary you built, run codesign -f -s - target/release/atlas. A browser download may also need xattr -d com.apple.quarantine ./atlas. |
atlas update says to run brew upgrade | The binary lives under a Homebrew prefix; update it with Homebrew. |
atlas update refuses to run | You are using a linked dev build (atlas link). Run atlas unlink first. |
| The update prompt is in the way | ATLAS_NO_UPDATE_PROMPT=1, or answer n (silences that version for 24 hours). |
atlas dev and the stack#
| Symptom | Cause and fix |
|---|---|
atlas dev needs an interactive terminal (exit 2) | stdin or stdout is not a TTY. From scripts and agents use atlas dev --status, --ci, --headless or --detach. |
atlas dev is already running (pid N) | One instance per project directory. Attach with atlas dev, check with atlas dev --status, stop with atlas down. |
atlas.yaml: services.web.helth: unknown field | A typo. Unknown keys are errors at every depth; the message suggests the right name. Run atlas config check. |
A service stays starting forever | It is waiting on its port, health URL or ready_log. The sidebar note says what (waiting for :3001 (12s)). After start_timeout (default 60 s) a warning is logged; the service keeps trying. Fix the probe or the port. |
A service is failed with dependency X failed | A dependency failed, exited non-zero or gave up restarting. Fix X first; its dependents start by themselves once their conditions hold, or restart them. |
gave up after N restarts | The restart budget (max_restarts, default 5) is spent. Read the service log, fix the cause, restart it from the menu (m) or with atlas_restart_service. |
| A service is green but not answering | A service without port, health or ready_log counts as running after 2 s alive. Add a probe. |
EADDRINUSE | Something else holds the port, often an orphan of an earlier run. Atlas cleans only what it started itself (from .logs/.pids) and never kills by port. Find the process with lsof -iTCP:<port> -sTCP:LISTEN and stop it yourself. |
docker compose failed and a docker alert shows | The daemon was not reachable within 15 s. Start Docker or OrbStack; app services still start. |
| Quitting leaves containers running | By design: Atlas never runs docker compose down. Stop them with docker compose down. |
pre_command failed toast | The dependency install failed; services start anyway. Read .logs/setup.log. |
Stale .logs/atlas.sock after a crash | Harmless; the next run replaces it. |
| Colors look wrong or text cannot be selected | M releases the mouse so your terminal can select; hold ⌥ while dragging otherwise. |
MCP and agents#
| Symptom | Cause and fix |
|---|---|
| The atlas tools are missing in the agent | atlas is not on the agent’s PATH (GUI apps often lack your shell PATH). Re-run atlas setup <agent> --apply from a shell where which atlas works so the absolute path is written. In omp: /mcp test atlas. |
atlas dev is not running; start it in a terminal | Control tools need a runner: atlas dev, atlas dev --detach or --headless. --ci does not serve control. |
atlas_status says running: false but the TUI is open | The agent is looking at a different project directory. Compare --dir or cwd. |
atlas_wait_ready is cut off after 30 s | The client timeout is shorter than timeout_s. Keep timeout_s ≤ 25 or raise the client timeout. |
The agent still runs atlas dev | The omp hook loads at session start: restart omp and check omp plugin list. /reload-plugins refreshes skills and MCP only. |
| Codex ignores the hooks | Trust them once with /hooks in Codex. |
CONFLICT … an atlas entry not created by atlas | You have your own atlas server entry. Rename it, or re-run with --force. |
The full table is in Agent setup and MCP.
Atlas Desktop#
| Symptom | Cause and fix |
|---|---|
atlas desktop cannot download the app | Check your connection and that the release for your CLI version exists. atlas desktop --status shows what is installed; atlas desktop --install --latest reinstalls. |
bun or node not found inside the app | An app opened from Finder gets a bare environment. Atlas reads your login shell’s environment at launch; Settings → General → Shell for your projects shows what it found and has Reload. Restart a running stack after changing it. |
| Terminals use the classic renderer with GPU on, and Settings says the terminal host is from an older Atlas | The daemon predates the app and could not be replaced live. Click Restart terminal host in Settings → Appearance → Terminal when no terminal is busy. |
| A terminal is blank right after opening | Turn off Fast terminal (GPU) in Settings → Appearance → Terminal to compare with the xterm.js renderer, and file an issue with atlas desktop debug logs output. |
| A worktree was not deleted with its branch | The branch is unmerged, so Atlas keeps it (a toast says so). Delete it with git branch -D when you are sure. |
atlas wt or atlas term says the app is not running | They use ~/.atlas/desktop.sock; open Atlas Desktop first. atlas term falls back to the local hostd when the app is closed. |
| The app and CLI versions differ | atlas update brings both to the latest; Update CLI & app in the app does the same. Running stacks keep their version until atlas down && atlas dev --detach. |
Debugging a debug build#
atlas desktop debug info
atlas desktop debug logs --level csp # CSP violations
atlas desktop debug shot /tmp/x.png
Reporting a bug#
If the CLI panics it prints a path such as crash-<timestamp>.toml under ~/Library/Application Support/atlas/crashes/ ($XDG_DATA_HOME/atlas/crashes on Linux). Crash reports stay on your machine and are never uploaded; review one for private data before attaching it.
Open an issue at github.com/codeatlasdev/atlas/issues with atlas version, your OS, what you ran and the relevant lines of .logs/<service>.log (redact secrets). Security problems go through SECURITY.md, not public issues.