Security risk report for @forgecat/voltagent_awesome-codex-subagents_research-analysis v0.0.14
Source Integrity
Low
Profile references a legitimate public GitHub repository (VoltAgent/awesome-codex-subagents) with clear attribution and MIT license.
No typosquatted, obfuscated, or attacker-controlled dependencies are introduced.
Agent Intent
Low
Content describes legitimate research and analysis agent workflows (competitive analysis, data research, documentation verification, market research, trend analysis) with no instructions to manipulate the AI, ignore system prompts, or exfiltrate information.
Agent instructions focus on quality checks, evidence sourcing, and decision-oriented synthesis—standard best practices for research tasks, not guidance poisoning or prompt injection.
No hidden instructions to weaken security defaults, install malicious dependencies, or reveal system prompts or conversation memory.
Details
Evidence
All agent descriptions emphasize source verification, confidence levels, and traceability (e.g., 'verify each comparison point is source-backed', 'verify answer statements map to concrete documentation references').
Sandbox mode is set to 'read-only' across all agents, limiting write/execute risk.
No references to external payload fetching, credential reading, or exfiltration channels.
Permissions
Low
No tools or hooks are declared in the profile; agents are pure instruction-based research workflows.
Sandbox mode 'read-only' restricts agents to information gathering and synthesis without file mutation, shell execution, or network write operations.
Authority is narrowly scoped to research and analysis tasks (searching, comparing, synthesizing) matching the stated profile purpose.
Details
Evidence
Declared tools section is empty.
All seven agents specify 'sandbox_mode: read-only', preventing high-risk operations.
Agent descriptions explicitly constrain scope (e.g., 'Do not make code changes or speculate beyond documentation evidence unless explicitly requested').
MCP Risk
Low
MCP server defined is 'openaiDeveloperDocs' pointing to a legitimate, well-known public documentation endpoint (https://developers.openai.com/mcp).
No arbitrary binary execution, unrestricted filesystem access, or hidden instructions in tool descriptions.
External host access is limited to a single, documented, read-only API endpoint.
Details
Evidence
MCP server URL is a standard OpenAI developer documentation resource, not an attacker-controlled or obfuscated endpoint.
No environment credentials or authentication tokens are required or declared.
No tool definitions with suspicious descriptions or hidden manipulation logic.
ForgeCat