feat: add MCode as a built-in ACP agent

[hetaoBackend]

Problem

MiniMax Code is an ACP v1 coding agent, but ACPX/OpenClaw users currently have to repeat a custom agent override before they can run it. This keeps the otherwise standard mcode acp entrypoint out of /acp spawn and the normal acpx <agent> workflow.

Proposed solution

Add one built-in agent profile:

mcode -> mcode acp

Keep the integration registry-only: no adapter package, protocol fork, provider credentials, or MCode-specific lifecycle logic in ACPX. MCode continues to own installation, login, models, permissions, and its embedded runtime.

Evidence

• Public package: @minimax-ai/code@0.1.2 on npm.
• The installed CLI advertises mcode acp as an ACP server over stdio.
• Its ACP implementation negotiates ACP v1 and implements initialize/authenticate, session creation, prompts, streaming updates, permissions, and cancellation.
• The launch contract is the same direct-binary pattern already used by built-ins such as gemini, cursor, kiro, pool, and zeroclaw.

Scope

• Add the synchronized string/argv registry entries and regression test.
• Add the neutral registry rows, agents/MCode.md, skill reference, and changelog entry required by repository policy.
• Validate against the published MCode CLI with a real ACPX session smoke.

The repository has no enabled GitHub Discussion categories, so this issue is the durable proposal requested by the contribution guide before the PR.

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed September 4, 2026, 4:56 AM ET / 08:56 UTC.

Summary
Keep open: neither current main nor v0.13.2 includes MCode, and the same-author implementation PR remains open. Its supplied real-agent transcript supports a concrete proposal, but built-in status still needs product approval.

Reproducibility: not applicable. this requests a new built-in profile, rather than reporting broken supported behavior; its absence is verified in current source and v0.13.2.

Ways to help us reproduce this

• Add a screenshot or short recording showing the behavior.
• Add expected vs actual behavior.

Maintainer decision needed

Question	Recommendation
Should acpx adopt mcode as a permanent built-in keyword despite the demonstrated agent version supporting only one-shot use across CLI invocations?	Approve the narrow built-in direction: Sponsor the linked PR's registry-only approach while retaining explicit one-shot limitations and ordinary PR validation.

Why: The native launch pattern fits existing profiles, but custom configuration already supplies named invocation and VISION.md requires deliberate acceptance of new conventions.

Next step
The open same-author implementation PR already owns this work; settle its product scope without opening a competing repair PR.

Review details

Best possible solution:

Resolve the built-in decision through #503, keeping any accepted integration registry-only with explicit session limitations and synchronized documentation.

Do we have a high-confidence way to reproduce the issue?

Not applicable: this requests a new built-in profile, rather than reporting broken supported behavior; its absence is verified in current source and v0.13.2.

Is this the best way to solve the issue?

Unclear whether built-in status is warranted: persistent custom configuration already supports named invocation, but a registry-only addition is the narrowest way to remove that setup.

AGENTS.md: found and applied where relevant.

Remaining risk / open question:

• The supplied MCode version does not advertise session reload; built-in support must not imply persistent conversation continuity that its one-shot evidence does not establish.

Codex review notes: model internal, reasoning high; reviewed against 95a9466b9b80.

Label changes

Label justifications:

• P3: This is an optional built-in convenience profile; existing custom-agent configuration already provides access to the native command.

Evidence reviewed

What I checked:

• Requested built-in remains absent: Both command registries omit mcode. Searches for MCode and MiniMax across source, tests, documentation, agent guides, skills, and changelog found no matching implementation. (src/agent-registry.ts:39, 95a9466b9b80)
• Latest release also lacks the profile: The v0.13.2 registry omits mcode and is unchanged relative to the reviewed main revision. (src/agent-registry.ts:39, fd173f04aa1b)
• Persistent custom configuration already exists: Global or project configuration accepts agents.<name>.argv. Configuring agents.mcode.argv as ["mcode", "acp"] avoids repeating a command override, while leaving the requested zero-configuration built-in distinction unresolved. (docs/CLI.md:419, 95a9466b9b80)
• Configured names enter the normal CLI: CLI registration includes configured agent names, and src/cli/flags.ts:559-565 preserves configured commands and argv before considering built-in defaults. (src/cli-public.ts:23, 95a9466b9b80)
• Open implementation and useful runtime evidence: The supplied GitHub snapshot identifies feat: add MCode as a built-in ACP agent #503 as an open, unmerged closing PR by the issue author. Its body includes an authenticated MCode 0.1.2 run returning ACPX_MCODE_OK and ACP output showing a completed tool call; it explicitly documents loadSession:false and one-shot scope. (5b40647f7cf4)
• Product convention requires an intentional choice: VISION.md favors ACP interoperability and a small core, but treats agent keywords as lasting API conventions whose convenience alone is insufficient. AGENTS.md also requires synchronized harness guides and skill documentation. (VISION.md:79, 95a9466b9b80)

Likely related people:

• Peter Steinberger: Suggested for follow-up; no historical authorship or introduction is verified. (role: unverified routing candidate; confidence: low)
• Onur Solmaz: Suggested for follow-up; no historical authorship or introduction is verified. (role: unverified routing candidate; confidence: low)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.