The Agent Skills Directory

[Privilege Escalation] (MEDIUM): The references/installation-guide.md file recommends using sudo apt install for installing clangd and ccls on Linux/Ubuntu systems. Although standard for system package management, it requires root privileges.
[Indirect Prompt Injection] (LOW): The skill's primary function involves reading and interpreting project-level configuration files (e.g., .lsp.json, tsconfig.json, pyrightconfig.json, compile_commands.json). This creates an attack surface where malicious instructions could be embedded in files provided by an untrusted project, though no active exploitation is present.
[Prompt Injection] (LOW): SKILL.md contains a 'MANDATORY' instruction block that attempts to strictly control the agent's behavior, forcing it to invoke another skill (docs-management) before responding to user queries. This is a common control-flow pattern but acts as a minor override of standard agent logic.
[External Downloads] (SAFE): All recommended download sources (Microsoft, Google, LLVM, Eclipse Foundation, etc.) and package registries (npm, PyPI, Go) are verified and trustworthy. The instructions do not use dangerous patterns like 'curl | bash'.

lsp-management