A curated collection of Zed coding agent skills, error references, and project rules. Each skill teaches the agent how to use a specific CLI tool effectively.
The repo itself is the deploy target — clone it directly into ~/:
git clone https://github.com/Prohect/.agents.git "$HOME/"If ~/.agents already exists, you may optionally combine them with this repo:
# Rename your old skills directory
mv "$HOME/.agents/skills" "$HOME/.agents-old/skills"
# Clone this repo into your home directory
git clone https://github.com/Prohect/.agents.git "$HOME/"
# combine them (optional if you already paid lots of effort maintaining your custom skills),
# you should read the contents recursively of both directories,
# and merge them manually to keep the rules not conflicting and confusing.
# Remove the old skills directory (optional)
rm -rf "$HOME/.agents-old".
├── README.md # You are here
├── AGENTS.md # Project agent rules for agents to maintain or contribute to this repository
├── Zed/
│ └── AGENTS.md # Global agent rule, you may override or combine with your old global AGENTS.md
│ # Recommand to use directory symlink to keep this one synced with your global AGENTS.md
├── skills/ # Agent skills (one per CLI tool)
│ ├── awk/ # GNU awk — text processing
│ ├── cat/ # GNU cat — file display
│ ├── commit-message/ # Git commit message writing
│ ├── errors/ # Tool error reference index
│ ├── es/ # Everything Search — instant file/directory search, only on Windows
│ ├── gh/ # GitHub CLI
│ ├── grep/ # GNU grep — content search
│ ├── ln/ # GNU ln + mklink — links & junctions, only on Windows
│ └── sed/ # GNU sed — stream editing
├── errors/ # Error solution docs (referenced by errors skill)
└── demo/ # Deterministic test fixtures for each skill
│ # Some subdirectories are .gitignored — they reference
│ # local files (crash dumps, browser profiles, etc.)
│ # that would leak private data or break on other machines.
│ # See each skill's SKILL.md for how to set up your own.
Each skill is a directory under skills/ containing a SKILL.md file with YAML frontmatter:
---
name: cat
description: Use GNU cat for reading, concatenating, and displaying files.
---
# cat — Concatenate & Display Files
...reference docs, examples, quirks, and edge cases...The agent loads a skill when a task matches its description. The SKILL.md teaches the agent the tool's full CLI surface — flags, quirks, platform notes, and worked examples — so it uses the tool correctly on the first try.
The errors skill is special: it's an index of documented tool-call errors and their solutions. When the agent hits a new error, it could consults errors/ for a known fix instead of guessing. Each error has its own directory with a SOLUTION.md.
Skills document the usage of these CLI tools, but you only need the ones you use:
| Tool | Minimum Version | How to get it |
|---|---|---|
| GNU awk | 5.3.0 | pacman -S gawk (MSYS2), or built-in on most Linux/macOS |
| GNU cat | 8.32 | pacman -S coreutils (MSYS2), or built-in |
| GNU grep | 3.0 | pacman -S grep (MSYS2), or built-in |
| GNU sed | 4.9 | pacman -S sed (MSYS2), or built-in |
| GNU ln | 8.32 | pacman -S coreutils (MSYS2), or built-in |
| Everything Search | 1.1.0.30 | voidtools.com (Windows only) |
| GitHub CLI | 2.95.0 | cli.github.com |
On Windows, install MSYS2 to get the GNU toolchain. The es and gh skills are Windows-first but the concepts transfer.
See AGENTS.md for the skill authoring workflow. TL;DR:
- Audit the tool (
<tool> --help,<tool> --version) - Build a demo directory under
demo/<skill>/ - Test every example against the demo before writing it
- Spawn a sub-agent to independently verify all examples
- Document quirks prominently (broken vs. working examples)
This is a personal tool configuration collection. Use, modify, and share freely.