The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes an anonymous telemetry module for tracking usage statistics.
Evidence: lib/telemetry.js transmits data such as event types and performance metrics to https://t.insightx.pro via fetch requests.
Context: The telemetry is disabled by default and requires the AGENT_REGISTRY_TELEMETRY environment variable to be set. It does not collect sensitive information such as search queries or agent names.
[EXTERNAL_DOWNLOADS]: The installer can optionally download the @clack/prompts package from the npm registry.
Evidence: install.sh performs a dependency installation if the --install-deps flag is used during setup.
[COMMAND_EXECUTION]: Multiple scripts are provided to perform file system operations on local agent directories.
Evidence: bin/init.js and bin/get.js facilitate copying, moving, and reading agent markdown files from standard Claude Code directories.
Security: Path resolution logic in lib/registry.js implements checks to prevent directory traversal and ensure operations stay within designated folders.
[PROMPT_INJECTION]: The skill uses a hook to process user input and suggest relevant agents.
Evidence: hooks/user_prompt_search.js analyzes user prompts to match them against the registry index and injects recommendations as additional context.
Analysis: This is an indirect injection surface where user input determines which local agent instructions are brought into focus.

agent-registry