The Agent Skills Directory

[COMMAND_EXECUTION]: The action exec-token allows the execution of arbitrary shell commands. The underlying script scripts/exec-token.js uses child_process.execSync with shell: true to run commands where a placeholder is replaced by a real API key fetched from the server.
[CREDENTIALS_UNSAFE]: The skill is designed to manage, inject, and copy sensitive sk- API keys. It also retrieves management credentials from local .env files using scripts/env.js and stores them in environment variables.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to external endpoints (user-configured NEWAPI_BASE_URL) to manage tokens and fetches documentation from apifox.newapi.ai and www.newapi.ai.
[DATA_EXFILTRATION]: The skill moves sensitive tokens between a remote API, local storage, the system clipboard, and shell command arguments. It relies on a regex-based sanitization script (scripts/sanitize.js) to redact these secrets from the AI's view, which is a heuristic that could potentially be bypassed or fail on unexpected formats.
[PROMPT_INJECTION]: The SKILL.md file contains extensive, "absolute and non-negotiable" instructions aimed at overriding the agent's default behavior to prevent it from revealing or inspecting the API keys it handles.
[DATA_EXPOSURE]: When applying a token to a configuration file, scripts/inject-key.js creates a .bak backup file in the same directory. This backup file contains the real, unmasked API key in plain text, posing a risk of accidental exposure if the file is not manually deleted.

newapi