browse-local-sentry
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
lsofto find local server ports and PIDs. It also usescurlandopensslto verify the presence and validity of SSL certificates on local dev servers. - [EXTERNAL_DOWNLOADS]: References the
chrome-devtools-mcpNode package and a specific browser extension from the Chrome Web Store to assist with session management. - [REMOTE_CODE_EXECUTION]: The skill instructs the user or agent to execute the
chrome-devtools-mcptool vianpx. This involves downloading and running code, which is performed with version pinning and explicit user consent. - [PERSISTENCE]: Provides shell commands to modify the agent's persistent MCP configuration (e.g.,
claude mcp add), ensuring the browser automation tools are available in future sessions. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from a local development server API.
- Ingestion points: Local server endpoint
/api/client-config/. - Boundary markers: None present in the skill's logic for processing the API response.
- Capability inventory: Shell command execution (
lsof,curl,openssl) and browser automation (list_pages). - Sanitization: No explicit sanitization of the server's JSON response is described, although its usage is restricted to an authentication check.
Audit Metadata