Windows Agent Shell is for a coding agent on Windows when output comes back garbled or something works on one machine and breaks on another. It records how Git Bash, PowerShell and cmd each split arguments, convert paths and decode text, with the version and date of every observation.
It ships two Python scripts: one prints what a program really received, the other flags mojibake, BOMs, CRLF and invalid UTF-8 in files.
What You Get
- Rules for slashes, drive letters and Git Bash rewriting of leading-slash arguments
- PowerShell traps: split flags, empty environment variables, native stderr and Get-Content encoding damage
- probe_argv.py, which prints the arguments, working directory and pipe encodings a program received
- check_text.py, which flags mojibake, BOM, CRLF and invalid UTF-8 and can fix CRLF
- A troubleshooting file with dated auto-mode workarounds
Common Use Cases
- gh api calls that fail because Git Bash rewrote a leading slash
- Accented or dash characters turning into garbage after a PowerShell edit
- Tests that fail on Windows because of CRLF or a lowercase drive letter
- Stopping a Windows background server by its process id or port
What Makes It Unique
The skill is built on measured behaviour. Each trap names the shell, the tool version and the date it was observed, and the agent probes the same arguments in the same shell to see what happened. It also covers the class where a third-party CLI differs on the target machine, such as branching on exit codes and encoding the older contract in test fakes.
Under The Hood
The skill is instructions, a troubleshooting reference and two standard-library Python scripts. The agent runs them in the shell: one only prints what it received, the other reads files and, with a fix flag, rewrites line endings in the files you name. The advice tells the agent to write commit and issue text to a file and pass it by path, to run long PowerShell scripts from the Bash tool, and to stop servers by process id or port. It does not fetch web pages.