The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install golangci-lint by piping a remote script from https://golangci-lint.run/install.sh directly into the shell (sh).
[EXTERNAL_DOWNLOADS]: The skill downloads and installs various development tools and packages from several sources:
Node.js packages: @biomejs/biome, oxlint, oxfmt, @colbymchenry/codegraph, chrome-devtools-mcp, playwright-cli, and @ast-grep/cli from the npm registry.
Python packages: ruff and ty from PyPI.
Binaries and tools via Homebrew: ast-grep, rtk, gh, glab, and golangci-lint.
Rust tools via Cargo: rtk from https://github.com/rtk-ai/rtk.
[COMMAND_EXECUTION]: The skill frequently uses subprocess.run and direct shell commands to perform its tasks:
scripts/inspect-project.py and scripts/guard-git.py execute Git commands to analyze repository state.
SKILL.md provides templates for running installation and configuration commands with output redirection to temporary logs.
Configuration of Git hooks using git config core.hooksPath.
[DYNAMIC_EXECUTION]: The skill dynamically generates a shell script for the Git pre-commit hook based on the detected project languages and tools, and writes it to the .githooks/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted data from the repository (such as package.json scripts, Git remotes, and file names) to decide which commands to execute and how to configure the environment.
Ingestion points: scripts/inspect-project.py reads project structure, configuration files (package.json, pyproject.toml), and tracked file names.
Capability inventory: The skill can execute package managers, modify Git configuration, and write executable scripts to the file system.
Boundary markers: No specific delimiters or safety instructions are used when interpolating project data into generated commands.
Sanitization: While the scripts use standard libraries for parsing structured data (JSON), there is no explicit sanitization of repository-derived strings before they are used in shell commands.
[DATA_EXFILTRATION]: The scripts/detect-harnesses.py script scans the user's home directory (~) and ~/.config to identify installed AI agent harnesses and their steering files (e.g., AGENTS.md, CLAUDE.md). While not exfiltrating data over the network, this represents access to files outside the project scope.

setup