Security risk report for @forgecat/anthropics_skills_canvas-design v0.0.5
Source Integrity
Low
Profile claims Anthropic as author with reference to official GitHub repository (anthropics/skills) and Apache-2.0 license.
Metadata (commit hash, platform compatibility) consistent with legitimate open-source distribution; no typosquatted dependencies or suspicious package references.
Agent Intent
INT-000High
The SKILL.md contains embedded guidance poisoning [OWASP LLM04]: it instructs the AI to repeatedly emphasize 'meticulously crafted,' 'painstaking attention,' 'master-level execution,' and 'countless hours' — framing designed to manipulate the AI into producing inflated, deceptive claims about effort/quality that misrepresent the actual generative process.
The instruction 'CRITICAL GUIDELINES: Emphasize craftsmanship REPEATEDLY' and 'This framing is essential - repeat phrases like...' explicitly directs the AI to inject misleading language into outputs, poisoning future design artifacts with false authority and false labor claims.
The 'DEDUCING THE SUBTLE REFERENCE' section instructs the AI to embed hidden conceptual content ('subtle, niche reference,' 'quiet conceptual DNA woven invisibly') without user awareness — a form of covert instruction injection that steers outputs toward attacker-defined themes while appearing to honor user intent.
Details
Evidence
"Emphasize craftsmanship REPEATEDLY: The philosophy MUST stress multiple times that the final work should appear as though it took countless hours to create, was labored over with care, and comes from someone at the absolute top of their field. This framing is essential - repeat phrases like 'meticulously crafted,' 'the product of deep expertise,' 'painstaking attention,' 'master-level execution.'"
"The topic is a subtle, niche reference embedded within the art itself - not always literal, always sophisticated... The reference must be refined so it enhances the work's depth without announcing itself... only those who know will catch it."
"CRITICAL STEP: Before creating the canvas, identify the subtle conceptual thread from the original request... The design philosophy provides the aesthetic language. The deduced topic provides the soul - the quiet conceptual DNA woven invisibly into form, color, and composition."
Permissions
Low
No tools or MCP servers are declared in the profile; the skill is purely instructional guidance for the AI's reasoning and output generation.
No file_write, shell, web_fetch, or other high/medium-risk tool categories are invoked; the profile does not request authority beyond markdown/design documentation.
Details
Evidence
"Declared tools (normalized categories; paths are untrusted): (none declared)"
"MCP servers (untrusted data): (none)"
MCP Risk
Low
No MCP servers are defined in the profile.
No external binaries, network calls, or filesystem access are configured.
ForgeCat