fix(cloud-agent): bump CLI to 7.4.20 and default bash timeout to 4m

[eshurakov]

Summary

Cloud Agent sessions were dying during heavy commands (like pnpm typecheck) with a generic wrapper failure, even though the sandbox was still alive.

This PR fixes two things that caused that:

1. Upgrade Kilo CLI to 7.4.20 — when a command is killed (OOM, SIGKILL, etc.), the agent now gets the exit code immediately and can continue, instead of hanging.
2. Default shell timeout of 4 minutes — if a command stalls with no output (e.g. thrashing at a memory limit), it is stopped before the 5.5-minute session liveness deadline, so the agent can recover instead of the whole session failing.

Also adds a small env passthrough so the default timeout can be overridden from Worker config if needed.

What we found

We reproduced the failure on a test Worker with a clean checkout of this monorepo:

Setup	Result
6 GB sandbox	Session dies during install/typecheck
12 GB sandbox	Session stays healthy; typecheck can finish
Command killed by signal (on 7.3.x)	Agent could hang waiting for exit
Same kill on 7.4.20	Exit 137 right away; agent continues
Long timeout + memory pressure	Process stays alive but stuck; session times out after ~5.5 min
Short timeout under same pressure	Command fails cleanly; agent runs a follow-up and finishes normally

So there were two separate issues:

• Killed process not reported → fixed by CLI 7.4.20
• Stuck-but-alive process → fixed by a 4-minute default shell timeout (under the 5.5-minute liveness limit)

This does not make a full monorepo typecheck fit in 6 GB — that still needs more memory. It makes failures visible and recoverable instead of taking down the session.

Test plan

• Unit tests for timeout default, override, and env filtering
• Deployed to Cloudflare test Worker (cloud-agent-next-evgeny-test)
• Confirmed kilo --version is 7.4.20 on that deployment
• Signal kill (kill -KILL $$) returns exit 137 immediately; agent runs a follow-up command
• Heavy stalled command with a short shell timeout fails cleanly; agent recovers and session completes normally
• Same heavy workload with a long timeout still hits the stuck-but-alive path (confirms why the default timeout is needed)

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

Reviewed the CLI 7.4.20 bump and KILO_EXPERIMENTAL_BASH_DEFAULT_TIMEOUT_MS=240000 hardcoded default with Worker-env passthrough: env-key allowlisting plus shell-quoting guard injection, override ordering is correct in both the process env record and the inlined command string, plumbing through ensureWrapper/ensureBootstrapWrapper into ensureRunning is intact, version pins are consistent across Dockerfiles, wrangler image_vars, package manifests, and the generated slash-command catalog, and the new tests cover default, override, and key-filtering behavior.

Files Reviewed (16 files)

• pnpm-lock.yaml
• services/cloud-agent-next/Dockerfile
• services/cloud-agent-next/Dockerfile.dev
• services/cloud-agent-next/Dockerfile.dind
• services/cloud-agent-next/package.json
• services/cloud-agent-next/src/agent-sandbox/cloudflare/cloudflare-agent-sandbox.test.ts
• services/cloud-agent-next/src/agent-sandbox/cloudflare/cloudflare-agent-sandbox.ts
• services/cloud-agent-next/src/kilo/devcontainer.ts
• services/cloud-agent-next/src/kilo/wrapper-client.test.ts
• services/cloud-agent-next/src/kilo/wrapper-client.ts
• services/cloud-agent-next/src/shared/default-slash-commands.generated.ts
• services/cloud-agent-next/src/shared/kilo-server-env.ts
• services/cloud-agent-next/src/types.ts
• services/cloud-agent-next/test/e2e/README.md
• services/cloud-agent-next/wrangler.jsonc
• services/cloud-agent-next/wrapper/package.json

---

_{Reviewed by kimi-k3 · Input: 70.3K · Output: 10.5K · Cached: 732.2K}

_{Review guidance: REVIEW.md from base branch main}