chore(runner): upgrade Claude ACP adapter

[mmabrouk]

Context

The runner still used the deprecated @zed-industries/claude-agent-acp 0.23.1 package. That release predates the adapter's native ACP effort configuration and no longer receives updates under the old package name.

Changes

• Replace the deprecated package with @agentclientprotocol/claude-agent-acp 0.58.1.
• Add the compatible Anthropic SDK peer at 0.111.0 and refresh the pnpm lock from current big-agents.
• Remove the stale secondary package-lock.json; runner Docker builds and CI use pnpm-lock.yaml with frozen installs.
• Update the runner's MCP transport comment to use the current package name.

Before this change, local Claude sessions launched adapter 0.23.1 and exposed no ACP effort option. After the change, the adapter reports effort in the thought_level category and accepts model-supported values through Session.setThoughtLevel().

Testing

• pnpm run typecheck
• 44 focused runner model, ACP, provider, and capability tests
• Isolated pnpm install --frozen-lockfile --ignore-scripts from current big-agents
• Rebuilt the EE dev runner image and recreated both the compose runner and loopback-only Claude subscription sidecar
• Verified adapter 0.58.1 in both rebuilt containers
• Verified http://144.76.237.122:8280/api/health and sidecar /health
• Created a Claude Sonnet ACP session, found { id: "effort", category: "thought_level" }, set low, and read back currentValue: "low"
• Ran a normal Claude/Sonnet /run through the rebuilt sidecar and received the exact expected response EFFORT-ADAPTER-OK

The full runner suite completed 826 of 829 tests. Three existing tests still fail reproducibly on the unchanged source: one permission-responder microtask timing assertion and two Pi transcript replays. They are outside this dependency-only change; focused ACP coverage and type checking pass.

Notes

This upgrades the adapter used by local runner sessions. Daytona uses the adapter baked into its sandbox image and must be upgraded and verified separately before claiming parity there.

Upstream issue #851 reports a background Claude subagent permission-channel deadlock. This PR verifies the basic session/config path; the broader permission, cancellation, resume, and subagent matrix remains follow-up QA.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agenta-documentation	Ready	Preview, Comment	Jul 10, 2026 11:16pm

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 4ef9c97c-c32f-40c2-a14d-1b5ff9b1058c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch chore/upgrade-claude-agent-acp

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[mmabrouk]

🤖 The AI agent says: CI triage: all runner checks are green, including unit, integration, and acceptance. The red SDK job is two assertions in unchanged agent permission/capability tests; the red API job is six SSRF-validation assertions in unchanged secrets tests. Neither job touches the four-file runner dependency diff, so I left those unrelated failures out of this PR.