ry-run/run-skills
Overview
This skill provides direct control over a local machine using Desktop Commander MCP. I use targeted, verifiable tool calls to read, search, edit files, manage Excel/PDFs, run terminal commands and REPLs, and inspect or terminate processes. The skill emphasizes small, safe steps, clear rollbacks, and explicit confirmation for high-risk actions.
How this skill works
I translate user requests into specific Desktop Commander MCP operations such as read_file, edit_block, write_file, start_process, interact_with_process, and read_process_output. Searches use start_search and pagination; edits use edit_block for precise replacements or chunked write_file for large rewrites. For interactive sessions I start a process, then stream input/output and manage sessions with list_sessions/list_processes.
When to use it
- You need to find files or specific content across a repo or filesystem.
- You want to view, tail, or page through large text files, logs, PDFs, images, or Excel sheets.
- You must make precise, minimal edits to code, configs, or spreadsheets.
- You need to run commands, launch REPLs (Python/Node/SSH/DB), or observe long-running jobs.
- You must inspect, debug, or safely terminate processes and review recent tool calls.
Best practices
- Prefer absolute paths and explicit offsets/lengths when reading files to avoid ambiguity.
- Read before you edit; use edit_block for targeted replacements and write_file in 25–30 line chunks for rewrites.
- Treat high-risk ops as requiring explicit confirmation: killing processes, config changes, bulk edits/moves, or destructive commands.
- Use start_search + get_more_search_results for repository exploration and stop_search when done to reduce noise.
- Remember allowedDirectories limits filesystem tools, not terminal commands—don’t assume it is a sandbox.
Example use cases
- Search a codebase for a function, open the file, apply a precise 1-line fix, and verify via read_file.
- Tail a 10k-line log with offset=-200, diagnose an error, and suggest a safe config change (request confirmation before applying).
- Launch a Python REPL to run a pandas analysis on a CSV, stream results back, and save a processed Excel range via edit_block.
- Open a PDF, replace or insert pages using the appropriate PDF write tool, and return a download link.
- List active sessions, inspect a runaway process’s output, then request confirmation to terminate or force_terminate it.
FAQ
I read the file first, then rewrite in controlled chunks using write_file (rewrite then append) with a recommended 25–30 line chunk size and keep a rollback plan.
What counts as high-risk and requires confirmation?
Config changes, killing or force-terminating processes, bulk file operations (moves/deletes), and any command that could cause data loss require explicit user confirmation before execution.