hackerone-brain-mcp-server
Fail
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions recommend using
npx -y h1-brain-mcpto set up the MCP server. This command automatically downloads and executes theh1-brain-mcppackage from the NPM registry without manual confirmation or integrity verification. - [REMOTE_CODE_EXECUTION]: By instructing the agent to run
npx -y h1-brain-mcp, the skill facilitates the execution of remote code. If the package name is targeted by typosquatting or if the third-party package is compromised, it could lead to arbitrary code execution on the host machine. - [COMMAND_EXECUTION]: The skill provides a TypeScript implementation (
scripts/generate-context.ts) intended to be run locally vianpx tsx. This script performs network operations and writes output to the console, representing local command execution controlled by the skill's logic. - [PROMPT_INJECTION]: The 'Creativity Directive' section includes instructions such as 'Think like an attacker. Adapt. Improvise.' While intended to guide the agent's persona for security research, such directives can potentially influence the agent to bypass standard operational constraints.
- [DATA_EXFILTRATION]: The skill manages the
H1_API_TOKENthrough environment variables. The provided code snippets and the recommended NPM package have full access to this sensitive credential and perform outbound network requests toapi.hackerone.com. While this is the intended functionality, it establishes a data flow for sensitive tokens to an external API. - [DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements an attack surface for indirect prompt injection. It fetches HackerOne program policies and disclosed reports (untrusted external data) and injects them directly into the agent's context using
JSON.stringifywithout sanitization or boundary markers. This could allow an attacker who controls a HackerOne program policy to influence the agent's behavior.
Recommendations
- AI detected serious security threats
Audit Metadata