fix: replace opencode with openclaw in prompt param conversion

[sweetmantech]

Summary

• Updates processCreateSandbox to convert prompt to openclaw agent --agent main --message instead of opencode run
• Updates MCP tool descriptions to reference OpenClaw instead of OpenCode
• Updates test to assert the new openclaw command format

Test plan

• Existing tests updated and passing (8/8)
• Test MCP run_sandbox_command tool with prompt param

🤖 Generated with Claude Code

Summary by CodeRabbit

• Chores
  • Updated sandbox command tool references from OpenCode to OpenClaw, including corresponding updates to command syntax and descriptions.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
recoup-api	Ready	Preview	Feb 20, 2026 0:11am

[coderabbitai]

📝 Walkthrough

Walkthrough

The pull request updates command references from OpenCode to OpenClaw across sandbox tooling, changing the command structure from "opencode run" to "openclaw agent --agent main --message" in both the tool registration and sandbox creation process. This is a textual configuration update with no functional logic changes.

Changes

Cohort / File(s)	Summary
Sandbox Tool Registration lib/mcp/tools/sandbox/registerRunSandboxCommandTool.ts	Updated prompt description and tool description to reflect OpenClaw branding and command syntax change from "opencode run" to "openclaw agent --agent main --message".
Sandbox Creation Process lib/sandbox/processCreateSandbox.ts	Updated command mapping to use openclaw agent command structure with --agent and --message flags instead of opencode run syntax.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Possibly related PRs

• feat: add prompt param to POST /api/sandboxes #231: Modifies the same prompt-to-command translation in processCreateSandbox.ts, establishing the initial opencode mapping that this PR now updates.
• feat: add run_sandbox_command and get_task_run_status MCP tools #224: Introduces the sandbox tool and processCreateSandbox code paths that these changes directly modify.

Poem

When code became a grasping Claw,
Commands evolved without a flaw,
OpenCode rests, its work is done,
OpenClaw's reign has just begun 🦅

🚥 Pre-merge checks | ✅ 1

✅ Passed checks (1 passed)

Check name	Status	Explanation
Solid & Clean Code	✅ Passed	PR maintains strong adherence to SOLID and Clean Code principles with proper separation of concerns, DRY implementation, and comprehensive test coverage.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch sweetmantech/myc-4296-mcp-create-sandbox-prompt-param

---

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 and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

lib/mcp/tools/sandbox/registerRunSandboxCommandTool.ts (1)

18-30: Mutual exclusivity of prompt and command is documented but not enforced.

The descriptions on both prompt (line 22) and command (line 15) say "Cannot be used with…", but the schema has no .refine() guard. A caller who passes both will silently have command/args overridden in processCreateSandbox without any error signal. Worth adding a Zod refinement here so the contract is both documented and enforced at the boundary.

♻️ Proposed refinement to enforce mutual exclusivity

-const runSandboxCommandSchema = z.object({
+const runSandboxCommandSchema = z
+  .object({
   command: z
     .string()
     .optional()
     .describe("The command to run in the sandbox. Cannot be used with prompt."),
   args: z.array(z.string()).optional().describe("Arguments for the command."),
   cwd: z.string().optional().describe("Working directory for the command."),
   prompt: z
     .string()
     .optional()
     .describe(
       'A prompt to pass to OpenClaw. Runs `openclaw agent --agent main --message "<prompt>"` in the sandbox. Cannot be used with command.',
     ),
   account_id: z
     .string()
     .optional()
     .describe(
       "The account ID to run the sandbox command for. Only applicable for organization API keys — org keys can target any account within their organization. Do not use with personal API keys.",
     ),
-});
+  })
+  .refine((data) => !(data.prompt && data.command), {
+    message: "prompt and command are mutually exclusive — provide one or the other, not both.",
+    path: ["prompt"],
+  });

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@lib/mcp/tools/sandbox/registerRunSandboxCommandTool.ts` around lines 18 - 30,
The schema documents that prompt and command/args are mutually exclusive but
doesn't enforce it, so add a Zod refinement on the root schema used in
registerRunSandboxCommandTool to reject inputs that include both prompt and
command/args; implement schema.refine(data => !(data.prompt && (data.command ||
(data.args && data.args.length > 0))), { message: "Cannot specify both prompt
and command/args", path: ["prompt","command"] }) (or equivalent) so callers
passing both prompt and command/args get a validation error instead of silent
override in processCreateSandbox.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@lib/mcp/tools/sandbox/registerRunSandboxCommandTool.ts`:
- Around line 18-30: The schema documents that prompt and command/args are
mutually exclusive but doesn't enforce it, so add a Zod refinement on the root
schema used in registerRunSandboxCommandTool to reject inputs that include both
prompt and command/args; implement schema.refine(data => !(data.prompt &&
(data.command || (data.args && data.args.length > 0))), { message: "Cannot
specify both prompt and command/args", path: ["prompt","command"] }) (or
equivalent) so callers passing both prompt and command/args get a validation
error instead of silent override in processCreateSandbox.