The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent or user to clone a source code repository from a non-trusted GitHub account: https://github.com/abundantbeing/hermes-browser-extension.git.
[REMOTE_CODE_EXECUTION]: The installation workflow involves running npm install and npm run build on the downloaded repository. This executes arbitrary build scripts and installs dependencies defined by a third party, which can lead to code execution from an unverified source.
[INDIRECT_PROMPT_INJECTION]: The extension captures active tab context (titles, URLs, readable text, headings, and metadata) and sends it to the Hermes Agent. This provides a vector for malicious instructions embedded in web pages to influence the agent's behavior.
Ingestion points: Content is scraped from active browser tabs via a content script and transmitted through the extension's side panel (SKILL.md).
Boundary markers: The documentation states that page text is "marked as external context" to mitigate accidental obedience, though this does not fully prevent adversarial injection.
Capability inventory: The captured data is sent to the Hermes Gateway/API, which may have access to tools, sessions, and memory across multiple skills.
Sanitization: There is no explicit description of sanitization, filtering, or validation of the captured web content before it is processed by the AI.
[COMMAND_EXECUTION]: The skill requires manual shell command execution for setup and verification, including hermes gateway run and cat dist/manifest.json | grep version.

hermes-browser-extension