The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches script templates and additional markdown instructions from the author's repository at github.com/Melvynx/ai-maxxing-skills. These resources are used to scaffold new tools.
[COMMAND_EXECUTION]: Executes system-level package managers including brew, npm, pipx, and uv to install official vendor CLIs as requested by the user. It also uses shell commands to verify installations and configure paths.
[DYNAMIC_EXECUTION]: Programmatically generates new Python scripts and markdown skill definitions (SKILL.md) at runtime, writing them to the user's home directory to create functional, persistent agent extensions.
[PRIVILEGE_ESCALATION]: The instructions include steps for using sudo when necessary for global package installations, though the skill explicitly directs the agent to obtain user confirmation first.
[PERSISTENCE]: Modifies local system configuration by creating symbolic links in ~/.claude/skills and instructing the user to add environment variables to their ~/.zshrc file to ensure generated tools remain available across sessions.
[INDIRECT_PROMPT_INJECTION]: A vulnerability surface exists because the skill processes external API documentation (provided via URL or text) to generate CLI code and agent instructions.
Ingestion points: External API documentation and URLs processed during Step 3b.
Boundary markers: No specific delimiters or "ignore instructions" warnings are applied when reading external documentation.
Capability inventory: High-privilege actions including tool installation (subprocesses), file system writes, and symlink creation are available after data ingestion.
Sanitization: The instructions attempt to mitigate risk by requiring the agent to use only "confirmed endpoints" and relying on a pre-defined script template.

connect-tool