The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from the repository being bootstrapped (specifically package.json, README.md, and existing source code) to automatically generate descriptions and configuration blocks in AGENTS.md.
Ingestion points: Reads package.json scripts and README.md content in the target directory (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are used when interpolating this external content into the generated documentation.
Capability inventory: Performs shell command execution (bun, git, bash) and file system writes across multiple scripts.
Sanitization: No explicit sanitization or validation of the ingested strings is performed before they are written to configuration files.
[COMMAND_EXECUTION]: The skill performs extensive shell operations to set up the environment.
Sensitive File Access: Attempts to modify Obsidian application configuration files located at ~/Library/Application Support/obsidian/obsidian.json (macOS) or ~/.config/obsidian/obsidian.json (Linux) to register new vaults.
Dynamic Loading: The startup.sh script executes other bash scripts from computed relative paths ($SCRIPT_DIR/../../docs-vault/scripts/init-vault.sh) to delegate tasks to other installed skills.
System Interaction: Uses pgrep to check for running processes and git for repository initialization.
[EXTERNAL_DOWNLOADS]: The skill uses bun create, bun add, and bunx to download and install project templates and tools (such as Next.js, Biome, Vitest, and Playwright) from the npm registry. These are standard operations for development tools using well-known services.

startup