The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install a CLI tool using uv tool install targeting a ZIP archive from a non-trusted GitHub repository: https://github.com/Panniantong/Agent-Reach/archive/refs/tags/v1.6.0.zip (SKILL.md).
[REMOTE_CODE_EXECUTION]: The agent-reach add <channel> command dynamically downloads and installs third-party dependencies and tools (e.g., bili-cli, boss-agent-cli, CloakBrowser) at runtime from various remote sources (SKILL.md, references/maintenance.md).
[DATA_EXFILTRATION]: The skill describes a session cloning feature that attempts to copy site-specific cookies from the user's primary Google Chrome profile to the skill's isolated browser environment if the 'OpenCLI' extension is detected (references/sessions.md). While intended for convenience, this involves accessing sensitive browser data.
[COMMAND_EXECUTION]: The skill requires the execution of multiple powerful shell commands for setup, maintenance, and operation, including agent-reach setup, agent-reach add, and agent-reach browser setup (SKILL.md, references/maintenance.md).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process unstructured data from arbitrary URLs and social media platforms, creating a surface for indirect prompt injection attacks.
Ingestion points: agent-reach read <url>, agent-reach <channel> search <query>, and YouTube transcript retrieval (SKILL.md).
Boundary markers: None identified in the provided documentation.
Capability inventory: The tool can execute shell commands for dependency management and browser control, and it can write data to an isolated profile directory (~/.agent-reach/) (SKILL.md, references/maintenance.md).
Sanitization: No specific sanitization or filtering of retrieved web content is mentioned before it is returned to the agent context.

agent-reach