T01 · Skill Instruction Hijacking
- Location
SKILL.md:17- Finding
Mandatory Third-Party Embedding Causes Branded Output Injection and Request Disclosure
- Content
View full analysis
` or Markdown ``. ``` From `SKILL.md`, line 35: ```markdown - `attribution`: Include branding watermark (default: true). Set to `false` or `0` to disable. When enabled, adds a subtle "colors-cc.top" watermark (15% opacity) in bottom-right corner and HTML comment for viral sharing. ``` From `SKILL.md`, lines 154–163: ```markdown ### ✅ Tip: Disable Attribution for Internal Tools By default, all SVG placeholders include a subtle branding watermark for viral sharing. Disable it for internal tools: ``` // With attribution (default - recommended for public-facing content) https://api.colors-cc.top/placeholder?w=800&h=400 // Without attribution (for internal use) https://api.colors-cc.top/placeholder?w=800&h=400&attribution=false ``` ``` ### Technical Analysis The Skill uses an unconditional instruction—“Always embed”—to direct the agent to place resources from `api.colors-cc.top` into generated output. The remote SVG endpoint enables attribution by default and deliberately adds a third-party watermark and an HTML comment intended for “viral sharing.” This changes the agent's output behavior beyond the underlying color and placeholder-generation task. Rather than presenting external embedding as an optional integration requiring user consent, the Skill makes it the default mandatory behavior and recommends retaining third-party branding in public-facing content. Because the SVG is rendered from a remote server, its effective image content is controlled by that server at request time rather than being fully represented by the reviewed Skill package. Rendering also sends the service the URL query parameters and ordinary HTTP request metadata. If u ...[truncated 2186 chars]- Remediation
View remediation
