The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes project manifest files to identify and suggest necessary CLI tools, which introduces a surface for indirect prompt injection if those files are maliciously crafted to influence the agent's installation decisions.
Ingestion points: Manifest files including package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod, Gemfile, and composer.json are processed by scripts/detect_project_type.sh.
Boundary markers: The skill does not explicitly describe delimiters or sanitization logic for the content of these project files before processing.
Capability inventory: The skill can execute shell commands, install software via system and language-specific package managers (apt, pip, npm, cargo, go, uv), and modify the user's PATH environment variable.
Sanitization: Not explicitly documented in the instruction set; the skill relies on internal scripts and structured parsing (e.g., jq) for catalog entries.
[COMMAND_EXECUTION]: The skill provides numerous instructions for executing shell commands to diagnose, install, and audit environment state, including the use of local scripts like install_tool.sh and check_environment.sh.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of tools from well-known external sources such as GitHub releases (e.g., ripgrep-all, qsv) and official package repositories.
[PERSISTENCE]: The instructions describe how to modify shell configuration files (~/.bashrc, ~/.zshrc) to ensure that newly installed tool paths are persisted across sessions.
[PRIVILEGE_ESCALATION]: The instructions mention the use of sudo in specific contexts, such as clearing the page cache for accurate benchmarking using hyperfine (echo 3 | sudo tee /proc/sys/vm/drop_caches).

cli-tools