Update to RelayDesk

{
  "description": "Run development work on a paired computer, server, or VM. Use when the user wants builds, tests, linters, Git or project CLIs run there, or wants remote repository failures investigated and fixed.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 642
    },
    {
      "relative_path": "assets/icon.png",
      "size_in_bytes": 3037
    }
  ],
  "name": "remote-development-workflow",
  "skill_md_contents": "---\r\nname: remote-development-workflow\r\ndescription: \"Run development work on a paired computer, server, or VM. Use when the user wants builds, tests, linters, Git or project CLIs run there, or wants remote repository failures investigated and fixed.\"\r\n---\r\n\r\n# Remote development workflow\r\n\r\nUse this skill when the user's coding, build, test, CLI, or repository task must run on a computer, server, or VM paired through RelayDesk.\r\n\r\n1. Establish the execution context.\r\n   - Reuse the target device and project path when already known.\r\n   - Otherwise call list_devices and locate the project with file, list, or search tools.\r\n   - Use an explicit working directory for project commands when possible.\r\n\r\n2. Inspect before changing.\r\n   - Read the smallest relevant files, logs, or configuration.\r\n   - Use targeted Git or project commands when they help establish state.\r\n   - Do not dump environment variables or credential stores to discover configuration.\r\n\r\n3. Execute the user's development workflow.\r\n   - Use start_process for builds, tests, linters, package scripts, Git diagnostics, compilers, or project CLIs.\r\n   - Follow long-running work with read_process_output.\r\n   - Use list_sessions before starting a duplicate job when prior conversation context suggests one may already be running.\r\n   - If ordinary non-secret interactive input is required, use interact_with_process; never enter passwords, tokens, OTPs, or other authentication secrets.\r\n\r\n4. Debug failures from evidence.\r\n   - Read the actual error output.\r\n   - Inspect the directly relevant files.\r\n   - Apply focused edits with file tools.\r\n   - Rerun the narrowest validation that proves or disproves the fix.\r\n   - Keep failed approaches visible as failed; do not report success from a file edit alone.\r\n\r\n5. Protect consequential workflows.\r\n   - Do not deploy, publish packages, push releases, rotate credentials, delete data, or perform destructive version-control operations unless the user clearly asked for that action.\r\n   - Do not weaken RelayDesk's command or filesystem safety boundary.\r\n   - If a tool is blocked by local policy, explain the boundary instead of bypassing it.\r\n\r\n6. Finish with execution evidence:\r\n   - target device and project,\r\n   - commands run,\r\n   - files changed,\r\n   - tests or builds and their outcomes,\r\n   - anything still failing or awaiting user input.\r\n\r\nPrefer making concrete progress over repeatedly re-analyzing the same unchanged state.\r\n"
}