Troubleshooting

Start with quick triage: confirm the binary and config (codewhale --version, ~/.codewhale/config.toml), enable verbose logs with RUST_LOG=deepseek_cli=debug when needed (RUST_LOG=deepseek_cli::client=debug for HTTP retries/reconnects), and capture the current state of ~/.codewhale/sessions and ~/.codewhale/tasks.

Turn hangs or the stream stops

If a foreground shell command is still running, press Ctrl+B to move it to the background (the turn keeps running and the command becomes a background job under /jobs); use Esc or Ctrl+C to cancel the turn itself. Inspect deepseek_cli::client retry logs and endpoint connectivity, and after a restart confirm the previously in-flight turn shows as interrupted rather than running.

Network outage / offline behavior

New prompts queue while offline, persisted to ~/.codewhale/sessions/checkpoints/offline_queue.json. Inspect with /queue list, restore connectivity, then re-send queued entries (/queue edit <n> plus Enter, or the normal input flow); the queue file clears when the queue empties.

Crash recovery

The checkpoint lives at ~/.codewhale/sessions/checkpoints/latest.json; startup begins a fresh session unless --resume/--continue is supplied. Resume explicitly with codewhale --resume <id> or Ctrl+R in the TUI; if the checkpoint schema is newer than the binary supports, upgrade the binary or remove the stale checkpoint.

Persistent state schema errors

Errors like schema vX is newer than supported vY affect sessions, runtime thread/turn/item records, and tasks. Confirm the binary version, back up the state directory before editing, then either run a newer compatible binary or archive the incompatible records and regenerate state.

MCP / tool execution failures

Validate the ~/.codewhale/mcp.json schema and server command paths, confirm the server process starts manually, and check sandbox denials in TUI history/logs. Use /mcp validate for diagnostics, temporarily disable a failing server to isolate the issue, and re-enable after verification.

Docker notes

Each release publishes a multi-arch Linux image to GitHub Container Registry. The default image is a conservative runtime image: it runs as the non-root codewhale user (UID/GID 1000:1000), grants no passwordless sudo, and keeps user state in a volume mounted at /home/codewhale/.codewhale. Pin a release tag instead of latest for reproducible installs.

docker volume create codewhale-home

docker run --rm -it \
  -e DEEPSEEK_API_KEY="your-api-key-here" \
  -v codewhale-home:/home/codewhale/.codewhale \
  -v "$PWD:/workspace" \
  -w /workspace \
  ghcr.io/hmbown/codewhale:latest

When a project needs apt-get, compiler toolchains, or package managers inside the container, do not change the default image contract — build an explicit toolbox image from docs/examples/Dockerfile.toolbox, and use one named state volume per project so sessions, config, and the offline queue do not bleed across workspaces. Never bake API keys or SSH private keys into custom images.

Source documents: docs/OPERATIONS_RUNBOOK.md, docs/DOCKER.md · Update docs-map.ts when changing.

Find the guidance you need.