wk-colima

Files (5)

• README.md
• SKILL.md
• 2026-08-05_stale-running-state.md · references/
• 2026-08-18_mise-managed-install-and-start.md · references/
• 2026-10-23_disk-full-prune.md · references/

SKILL.md (first lines)

# Colima

Ensure Colima is running and healthy before any `docker` or `colima` command. Also run on `colima
start|stop|restart|status`, Docker daemon errors (`Cannot connect to the Docker daemon`, `Error response from
daemon`), a missing/unresponsive Docker socket, or build/run/compose failures where the daemon is the suspect.
Requires `colima` via mise (`mise use -g colima@latest`; Lima installs alongside it, not as a separate package), the
`docker` CLI, and `nproc` or `sysctl` (macOS fallback).

Explicit `/wk-colima`: `status` → Step 1 only, report and exit; `start` → Steps 1–3; `stop` → `colima stop` only;
`restart` → Step 4 unconditionally.

## Step 1: Check status

```bash
mise exec -- colima status 2>&1
```

Take the first that matches: 1) `colima` not installed → stop and report it must be installed
(`mise use -g colima@latest`); 2) `Running` → healthy, proceed, skip Steps 2–3; 3) `Stopped`, `not found`, or any
error → Step 2.

## Step 2: Compute CPU and memory limits

Derive CPU and memory from the host and halve both (floor). Never hardcode memory — the VZ driver rejects requests
above the host's `maximumAllowedMemorySize`. Disk (100 GB) stays constant.

```bash
PROC=$(nproc 2>/dev/null || sysctl -n hw.logicalcpu 2>/dev/null || echo 8)
CPU=$(( PROC / 2 ))
[ "$CPU" -lt 1 ] && CPU=1
MEM_BYTES=$(sysctl -n hw.memsize 2>/dev/null || echo 17179869184)
MEM_GB=$(( MEM_BYTES / 1073741824 / 2 ))
[ "$MEM_GB" -lt 1 ] && MEM_GB=1
echo "Starting colima with CPU=$CPU / ${MEM_GB} GB / 100 GB"
```

## Step 3: Start Colima

```bash
mise exec -- colima start --cpu "$CPU" --memory "$MEM_GB" --disk 100 --mount-inotify --very-verbose
```

Wait for exit; zero means started. Confirm Docker is reachable — resolve the socket from `docker context ls`, never
assume `$HOME/.colima/` (a mise-managed colima serves under `$HOME/.config/colima/`):

```bash
docker context ls
docker info > /dev/null 2>&1 && echo "Docker OK" || echo "Docker not reachable"
```

**Important:** `colima start` claims the VM is already running while `colima status` or `docker info` disagrees →
treat runtime state as stale and go to the forced-stop restart sequence (Step 4); do not retry start in place.

## Step 4: Restart sequence (Colima or Docker is broken)

Use when `colima start` exits non-zero, Docker is unresponsive after a start, any Docker command returns `Cannot
connect to the Docker daemon` or `Error response from daemon` during an otherwise normal session, or the user says
"colima is broken", "restart colima", or "docker isn't working". **Full shutdown first — never skip this step.**

```bash
# 1. Stop containers gracefully (best-effort — don't block on failure)
docker ps -q 2>/dev/null | xargs -r docker stop 2>/dev/null || true

# 2. Hard-stop Colima
mise exec -- colima stop --force 2>/dev/null || true

# 3. Wait for shutdown (socket may linger)
sleep 3

# 4. Start fresh
PROC=$(nproc 2>/dev/null || sysctl -n hw.logicalcpu 2>/dev/null || echo 8)
CPU=$(( PROC / 2 ))
[ "$CPU" -lt 1 ] && CPU=1
MEM_BYTES=$(sysctl -n hw.memsize 2>/dev/null || echo 17179869184)
MEM_GB=$(( MEM_BYTES / 1073741824 / 2 ))
[ "$MEM_GB" -lt 1 ] && MEM_GB=1
mise exec -- colima start --cpu "$CPU" --memory "$MEM_GB" --disk 100 --mount-inotify --very-verbose
```

After restart, re-run `docker info` to confirm the daemon is reachable. Restart sequence fails twice consecutively →
report the full `colima start` output to the user; a VM-level issue may need manual intervention (e.g., `colima
delete` to wipe state and start from scratch).

## Disk full: prune build cache and images, never volumes

Build fails with `not enough free space`, `no space left on device`, or a package manager out-of-space error → the
fixed-size VM disk is full; a restart reclaims nothing.

```bash
docker system df
docker builder prune -af
docker image prune -f
```

- Confirm via `docker system df` that build cache or images hold the reclaimable space, prune, then retry the build.
- **Never prune volumes** (`docker volume prune`, `docker system prune --volumes`) — they hold DB data shared across
  worktrees.

## Step 5: Report state

After any start or restart, emit: "Colima running: CPU={n}, memory={m} GB, disk=100 GB. Docker reachable." Step 1
found it healthy → emit nothing.

## Post-Completion

Invoke `wk-learn colima`.

From whizzzkid/skills, MIT.

More from whizzzkid/skills