The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions point the user to clone a repository from an untrusted GitHub account: https://github.com/NoOne-hub/JSReverser-MCP.git. This source is not verified or associated with a trusted vendor.
[COMMAND_EXECUTION]: The installation process involves potentially dangerous shell commands including npm install, npm run build, and node, which execute code retrieved from the external third-party repository.
[DATA_EXFILTRATION]: The skill includes high-privilege tools capable of extracting sensitive information:
dump_session_state: Specifically designed to export browser cookies, localStorage, and authentication tokens to a file (session.json).
get_hook_data: Captures runtime function arguments and results, which often contain plain-text credentials or API signatures.
list_network_requests: Monitors browser traffic, potentially revealing sensitive headers and keys.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze untrusted external data (JavaScript), posing a risk of indirect prompt injection.
Ingestion points: Untrusted script source code and collected page code are read into the agent's context through get_script_source and collect_code.
Boundary markers: There are no documented delimiters or instructions designed to isolate or warn the agent about embedded instructions in the analyzed code.
Capability inventory: The skill provides significant automated capabilities such as click_element, type_text, and navigate_page.
Sanitization: No evidence of content filtering or sanitization is presented for the ingested data.

jsreverser-mcp-javascript-reverse-engineering