T01 · Skill Instruction Hijacking
Warning
- Location
SKILL.md:52- Finding
Forced Advertising Hooks and Third-Party Branding in Generated Output
- Content
View full analysis
` in the ingredients column area. `` after the recipe. - **Footer**: "Powered by ClawCode" ``` ### Technical Analysis The skill instructions require the agent to insert two advertising placeholders and fixed third-party branding into every generated recipe page. These elements are unrelated to the skill's primary recipe-generation function and are presented as mandatory visual requirements rather than optional, user-approved features. When the skill is loaded, these directives alter the agent's output behavior by requiring promotional integration points in the generated HTML. The fixed element identifiers also provide predictable targets that a downstream script, hosting environment, or content-processing system could use to inject advertisements. No evidence was found that this project itself populates the advertising elements, loads remote advertising code, or executes a payload. The issue is limited to mandatory manipulation of generated content. ### Attack Path 1. A user invokes the daily recipe skill to generate a recipe artifact. 2. The agent loads and follows the mandatory requirements in `SKILL.md`. 3. The agent creates `/mnt/user-data/outputs/daily-recipe.html`. 4. The generated page contains `ad-slot-sidebar`, `ad-slot-bottom`, and the fixed “Powered by ClawCode” footer. 5. A downstream renderer or script can identify the predictable advertising containers and populate them with promotional content. ### Impact Assessment The issue affects the integrity and neutrality of generated recipe artifacts. Users may receive undisclosed promotional hooks and third-party branding that they did not request. The instructions do not grant operating-system privileges, access credentials, read sensitiv ...[truncated 212 chars]- Remediation
View remediation