Security risk report for @forgecat/mango_skills_agent-skills-game-development v0.1.1
Source Integrity
Low
The profile is authored by Meng To and sourced from a public GitHub repository (`https://github.com/MengTo/Skills/tree/main/agent-skills/game-development`) with MIT license and a specific commit hash.
No supply-chain manipulation, typosquatted dependencies, or attacker-controlled package recommendations are present.
All referenced tools and libraries (Three.js, React, WebGL) are well-known, legitimate, and standard for browser game development.
Agent Intent
Low
The profile contains no instructions to ignore system prompts, read credentials, exfiltrate data, or install remote payloads.
Content is purely educational and prescriptive for game development workflows (level design, audio, camera, inventory, combat, etc.)—it describes best practices and architectural patterns, not malicious actions.
The map-editor skill includes security guidance (IP allowlist, password hashing, session signing, secure cookies) that *strengthens* security rather than weakening it; this is defensive documentation, not poisoning.
Details
Evidence
Example: 'Treat architecture as gameplay communication. Every route, arena, gate, prop, and light must help the player read movement, threats, objectives, or state.' — describes level design intent, not manipulation.
Example: 'Keep passwords, allowlists, and signing material in deployment secrets; never bundle, log, export, or test-snapshot them.' — explicitly instructs secure handling of sensitive data.
No paraphrased or obfuscated instructions to manipulate the AI, leak system prompts, or perform unauthorized actions.
Permissions
Low
No tools or MCP servers are declared in the profile; it is a collection of markdown skill descriptions and reference architectures.
The skills describe game development workflows (authoring levels, building UI, tuning AI) that do not require shell, file-write, file-delete, or other high-risk tool categories.
Permissions are implicitly scoped to the game development domain and do not request authority beyond the stated function of each skill.
Details
Evidence
Profile declares: '(none declared)' for tools and MCP servers.
Skills reference only standard game-development libraries and patterns (Three.js, React, WebGL, TypeScript) without requesting elevated system access.
No `alwaysApply=true` rules, glob patterns, or excessive agency claims.
MCP Risk
Low
No MCP servers are defined in the profile.
The profile is purely markdown documentation and reference code; it does not invoke external binaries, network services, or arbitrary code execution.
Reference implementations (e.g., Vesperfall map-editor) are pseudocode and architectural guidance, not executable payloads.
Details
Evidence
Profile declares: '(none)' for MCP servers.
All code examples are TypeScript/JavaScript pseudocode with clear security boundaries (e.g., session signing, IP validation, secure cookies).
No hidden instructions in tool descriptions or unknown binary execution paths.
ForgeCat