fix(ci): build.yml runs out of disk on GitHub-hosted runners

[scottschreckengaust]

Problem

The build job in build.yml fails with ENOSPC: no space left on device during CDK test/synth esbuild bundling. This affects all PRs (e.g., #144 — a docs-only PR).

Error: Failed to write to output file: write /tmp/cdk.outRhUxmE/bundling-temp-.../index.js: no space left on device

Root cause

The build runs on ubuntu-latest runners (~25-30 GB free). Disk is consumed by:

Phase	Approximate disk used
Pre-installed software (Android SDK, .NET, tools)	~55 GB (only ~25-30 GB remains)
yarn install (node_modules, 3 workspaces)	~1.5 GB
uv sync (agent .venv)	~200 MB
//cdk:test → Jest triggers CDK synth → esbuild bundles ALL Lambda handlers to /tmp	~2-4 GB
//cdk:synth:quiet → esbuild bundles ALL handlers AGAIN to /tmp + writes cdk.out/	~2-4 GB

The CDK test step and CDK synth step both do full esbuild bundling independently, accumulating artifacts in /tmp.

Recommended fix (Option 1 — preferred, ~10 seconds added to build)

Add a disk cleanup step before Install dependencies:

- name: Free disk space
  run: sudo rm -rf /usr/local/lib/android /usr/share/dotnet /opt/hostedtoolcache

This frees ~22 GB in under 10 seconds by removing pre-installed Android SDK (~14 GB), .NET (~2.7 GB), and tool cache (~5.9 GB). No apt-get (which is what makes full cleanup actions take 5+ minutes).

For maximum headroom (~28 GB, still under 30 seconds):

- name: Free disk space
  run: |
    sudo rm -rf /usr/local/lib/android /usr/share/dotnet /opt/hostedtoolcache \
      /opt/ghc /usr/local/.ghcup /usr/local/share/powershell
    sudo docker image prune --all --force

Alternative fixes (lower priority)

Option 2: Clean /tmp between CDK test and synth

Add rm -rf /tmp/cdk.out* after //cdk:test and before //cdk:synth. This addresses the double-bundling accumulation but doesn't fix the underlying tight headroom.

Could be done in cdk/mise.toml:

[tasks."synth:quiet"]
description = "cdk synth (quiet, with pre-clean)"
run = ["rm -rf /tmp/cdk.out*", "npx cdk synth -q"]

Option 3: Use jlumbroso/free-disk-space action (MIT license)

- uses: jlumbroso/free-disk-space@main
  with:
    large-packages: false  # Skip apt-get (saves 3-5 min)
    android: true
    dotnet: true
    haskell: true
    tool-cache: true
    docker-images: false   # We need Docker for CDK bundling
    swap-storage: false

This takes ~30-60 seconds with large-packages: false. With it enabled, 3-5 minutes.

FOSS actions evaluated

Action	License	Speed	Notes
jlumbroso/free-disk-space	MIT	30s–8min	Best option if using an action; disable large-packages for speed
easimon/maximize-build-space	MIT	Minutes	Also does LVM resize — overkill for our need
Inline rm -rf	N/A	<10s	Simplest, fastest, zero dependencies

References

• PR docs: ADR-005 through ADR-013 (feedback, flags, knowledge, DoD, security, recovery, conflicts, knowledge stack, validation pyramid) #144 build failure: https://github.com/aws-samples/sample-autonomous-cloud-coding-agents/actions/runs/26475731407/job/77960682403
• GitHub runner specs: ~87 GB total, ~25-30 GB free after pre-installed software

[krokoko]

this was fixed by #190 , closing will reopen as needed