The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to manage git repositories and documentation workspaces. Evidence in workflow/connect-mode.md shows usage of git submodule commands to integrate external repositories. workflow/create-mode.md involves git init, git add, and git commit to initialize documentation workspaces.
[EXTERNAL_DOWNLOADS]: The skill performs automated package installation from the well-known NPM registry during workspace setup. Evidence in workflow/create-mode.md shows npm install --silent is called within the scripts/ directory to fetch the pdf-parse dependency.
[CREDENTIALS_UNSAFE]: The skill accesses its own local configuration files containing authentication data for git operations. Evidence in evals/evals.json and workflow/update-mode.md indicates it reads ~/.config/agent-wiki/credentials to synchronize documentation with remote servers.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by processing untrusted external data to generate documentation. 1. Ingestion points: Processes .pdf, .txt, and .md files as specified in workflow/create-mode.md. 2. Boundary markers: The workflows do not define specific delimiters or instructions to ignore embedded prompts in processed files. 3. Capability inventory: Includes file system write operations, git repository manipulation, and execution of local Node.js scripts. 4. Sanitization: No explicit content validation or sanitization is described before the extracted text is used for document generation.

agent-wiki