feat: E2E Docker testing + CLI tool sandbox test audit

[NathanFlurry]

Summary

• E2E Docker testing infrastructure (US-001 through US-008): Adds Docker-backed integration tests that spin up real Postgres, MySQL, Redis, and SSH/SFTP containers and run fixture operations from inside the secure-exec sandbox. All fixtures currently expect failure (net bridge is deferred) but serve as regression gates for when net is implemented.
• CLI tool test audit: Adversarial review found that all 6 cli-tools/ test files (Pi, Claude Code, OpenCode — headless + interactive) completely bypass the secure-exec sandbox, spawning tools directly on the host. New PRD (ralph/cli-tool-sandbox-tests) tracks rewriting them to run through the sandbox properly.
• Archives completed kernel-hardening and e2e-docker-testing PRDs.

What was added

• docs-internal/specs/docker-testing.md — full spec for Docker-backed e2e testing
• packages/secure-exec/tests/e2e-docker.test.ts — test runner with container lifecycle
• 5 fixture projects: pg-connect, mysql2-connect, ioredis-connect, ssh2-connect, ssh2-sftp-transfer
• packages/secure-exec/tests/e2e-docker/dockerfiles/sshd.Dockerfile — custom SSH test image
• packages/secure-exec/tests/utils/docker.ts — buildImage() helper
• .github/workflows/e2e-docker.yml — CI workflow with GitHub Actions services

CLI tool sandbox test findings

File	Status	What it does instead
pi-headless.test.ts	Bypasses sandbox	Host spawn(node, [pi-cli])
pi-interactive.test.ts	Bypasses sandbox	Host PTY via script -qefc
opencode-headless.test.ts	Bypasses sandbox	Host spawn('opencode') + host SDK client
opencode-interactive.test.ts	Bypasses sandbox	Host PTY via script -qefc
claude-headless.test.ts	Bypasses sandbox	Host spawn(claude)
claude-interactive.test.ts	Bypasses sandbox	Host PTY via script -qefc

All other tests (test-suite, runtime-driver, project-matrix, etc.) are clean — they properly use NodeRuntime/proc.exec().

Test plan

• E2E Docker tests skip gracefully without Docker (skipUnlessDocker())
• E2E Docker tests discover and run fixtures when Docker is available
• All fixture parity tests pass for host Node.js runs
• All fixture sandbox runs fail with expected net.* errors (net bridge deferred)
• CI workflow runs successfully with GitHub Actions services
• Existing tests unaffected

🤖 Generated with Claude Code

[pkg-pr-new]

Open in StackBlitz

secure-exec

pnpm add https://pkg.pr.new/rivet-dev/secure-exec@14

@secure-exec/browser

pnpm add https://pkg.pr.new/rivet-dev/secure-exec/@secure-exec/browser@14

@secure-exec/core

pnpm add https://pkg.pr.new/rivet-dev/secure-exec/@secure-exec/core@14

@secure-exec/node

pnpm add https://pkg.pr.new/rivet-dev/secure-exec/@secure-exec/node@14

@secure-exec/python

pnpm add https://pkg.pr.new/rivet-dev/secure-exec/@secure-exec/python@14

@secure-exec/typescript

pnpm add https://pkg.pr.new/rivet-dev/secure-exec/@secure-exec/typescript@14

commit: 3a3ab2a