The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the tool by piping a script from a remote URL directly to a shell: curl -fsSL https://raw.githubusercontent.com/nashsu/autocli/main/scripts/install.sh | sh. This allows for the execution of unverified code from an external repository without a prior integrity check.- [PERSISTENCE]: The documentation provides commands to append content to shell initialization files such as ~/.bashrc and ~/.zshrc for shell completions. While common for CLI tools, this represents a persistent modification to the user's environment.- [PRIVILEGE_ESCALATION]: Manual installation instructions require the use of sudo to move the downloaded binary into system-level directories (/usr/local/bin/), which demands administrative privileges.- [INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to scrape data from social media and external websites (e.g., Twitter, Reddit, Bilibili) and present it to an AI agent. This creates a surface for indirect prompt injection where untrusted external content could influence the agent's behavior. Evidence includes ingestion from over 55 sites via commands like autocli reddit search. No explicit sanitization or boundary markers are defined for the agent's context.- [CREDENTIALS_UNSAFE]: The tool manages and stores sensitive information, including API tokens and browser session data, in a local configuration file at ~/.autocli/config.json.- [COMMAND_EXECUTION]: The autocli register command allows the AI agent to access and execute high-privilege CLI tools such as gh, docker, and kubectl, significantly expanding the agent's capability to modify the user's infrastructure.

autocli-web-scraping