The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically triggers package managers (pnpm install or npm install) to download and install dependencies at runtime based on the presence of a package.json file.
[COMMAND_EXECUTION]: The skill executes broad file system searches (find) starting from the home directory (~/) and sensitive locations like ~/Documents, ~/Desktop, and iCloud folders to locate personal Obsidian vaults. It also performs bulk file operations, including copying entire vault structures (cp -r) and modifying repository metadata (rm -rf .github).
[COMMAND_EXECUTION]: The skill modifies system configuration files by instructing the agent to add environment variables (API keys) to shell profiles such as .zshrc and .bashrc.
[CREDENTIALS_UNSAFE]: The skill guides the user to store sensitive API keys for Gemini and Firecrawl directly in shell configuration files or .mcp.json, which may lead to insecure credential management if not handled carefully by the user.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it is designed to ingest and process untrusted external data from the user's Obsidian vault. Specifically, it samples 10-15 random notes and lists numerous filenames to detect patterns, which could contain malicious instructions designed to hijack the agent's behavior.
Ingestion points: Reads random markdown files from the user-provided Obsidian vault path and analyzes .obsidian/community-plugins.json (SKILL.md).
Boundary markers: None identified in the instructions for the sampled content.
Capability inventory: The skill has access to Bash, Write, MultiEdit, and Task tools, allowing for significant system modification (SKILL.md).
Sanitization: No evidence of sanitization or filtering of the sampled note content before processing.

init-bootstrap