heating the forge...

@forgecat/msitarzewski_agency-agents_agency-game-development — Security Risk Report | ForgeCat

msitarzewski_agency-agents_agency-game-development

Security risk report for @forgecat/msitarzewski_agency-agents_agency-game-development v0.0.11

Low

Profile declares origin as msitarzewski/agency-agents (GitHub), with MIT license and specific commit hash (9c31d86, 2026-03-23).
Static scanner confirms safe level — no supply-chain red flags detected.
Content is consistent with a legitimate open-source game development agent collection.

Low

Content is educational documentation of game development specializations (Blender, audio, design, Godot, Unreal, Unity, etc.) with no instructions to manipulate the AI, ignore system prompts, or perform malicious actions.
Markdown describes agent personalities, technical workflows, and code examples — all framed as guidance for legitimate game development tasks, not as hidden instructions to the AI.
No prompt injection, role hijacking, credential exfiltration, system prompt leakage, or guidance poisoning detected. Code examples (Python, GDScript, C#, GLSL) are standard best practices for their respective domains, not backdoored templates or security-weakening guidance.

Details

Findings

Invisible characters detected
agents/level-designer.md · SKILL-011
Invisible characters detected
agents/unity/unity-architect.md · SKILL-011
Invisible characters detected
agents/unity/unity-editor-tool-developer.md · SKILL-011
Invisible characters detected
agents/unreal-engine/unreal-systems-engineer.md · SKILL-011

Evidence

Example: 'You are **BlenderAddonEngineer**, a Blender tooling specialist who treats every repetitive artist task as a bug waiting to be automated.' — describes role, does not instruct AI to bypass controls.
Example: 'MANDATORY: Prefer data API access (`bpy.data`, `bpy.types`, direct property edits) over fragile context-dependent `bpy.ops` calls' — standard Blender API guidance, not malicious.
Example: 'Every tool must save time or prevent a real class of handoff error' — legitimate design principle, not an instruction to weaken security or hide behavior.

Low

Profile declares no tools, hooks, or MCP servers — it is pure markdown documentation of agent personalities and technical guidance.
No file_write, file_delete, shell, web_fetch, or subagent permissions are requested or implied.
Authority is implicitly limited to the scope of each agent's stated domain (Blender scripting, audio integration, game design, etc.) with no excessive or unrelated capabilities.

Details

Evidence

Profile metadata: 'Dependencies: None' and 'MCP servers: (none)' — no external tool integration.
Content is instructional (code examples, design patterns, workflow guidance) not executable — no runtime permissions required.

Low

No MCP servers are defined in the profile.
No tool descriptions, binary execution, network access, or filesystem permissions are declared.
Profile is pure markdown documentation with no server-side or external integration.

Details

Evidence