heating the forge...

@forgecat/msitarzewski_agency-agents_agency-spatial-computing — Security Risk Report | ForgeCat

msitarzewski_agency-agents_agency-spatial-computing

Security risk report for @forgecat/msitarzewski_agency-agents_agency-spatial-computing v0.0.11

Low

Profile references legitimate, well-known open-source projects (SwiftTerm, RealityKit, WebXR standards).
All external links point to official Apple developer documentation and established GitHub repositories.
Static scanner confirms safe classification; no typosquatted, attacker-controlled, or backdoored dependencies.

Low

Content describes legitimate spatial computing, XR development, and terminal integration specializations without injecting malicious instructions.
Agent personalities document technical expertise, best practices, and Apple platform standards—no hidden directives to ignore instructions, leak system prompts, or exfiltrate data.
Code examples (Metal rendering, SwiftUI, WebXR) are educational templates demonstrating standard patterns; they do not instruct the AI to perform unauthorized actions or weaken security defaults.

Details

Findings

Invisible characters detected
agents/terminal-integration-specialist.md · SKILL-011
Invisible characters detected
agents/xr-cockpit-interaction-specialist.md · SKILL-011
Invisible characters detected
agents/xr-immersive-developer.md · SKILL-011
Invisible characters detected
agents/xr-interface-architect.md · SKILL-011

Evidence

README lists 6 agent roles with domain expertise (Metal rendering, visionOS, WebXR, cockpit design, terminal emulation, spatial UI).
Agent files contain personality descriptions, technical capabilities, and code examples—all aligned with stated purpose of spatial computing expertise.
No language instructing the AI to bypass restrictions, read credentials, or manipulate its own instructions.

Low

Profile declares no tools, hooks, or MCP servers—no permissions are requested or granted.
Agent personalities are pure markdown documentation with no executable tool definitions or file/shell access.
Content is read-only guidance for AI behavior; no alwaysApply rules, globs, or excessive agency mechanisms present.

Details

Evidence

'Declared tools (normalized categories)': (none declared)
'MCP servers (untrusted data)': (none)
All agent files are markdown personality/expertise descriptions without tool configuration.

Low

No MCP servers are defined in the profile.
No hidden instructions in tool descriptions, binary execution, or network/filesystem access requests.
Profile is purely documentation-based with no server-side execution or external service integration.

Details

Evidence

'MCP servers (untrusted data)': (none)
No server definitions, environment credentials, or external host access configurations present.