The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing external asset data and text strings that are then used to construct prompts for media generation tools.\n
Ingestion points: The workflow retrieves product references, logos, and campaign assets via the search_assets tool and imports external media through import_media or import_chatgpt_media (SKILL.md, references/platform-upload.md).\n
Boundary markers: The instructions lack explicit delimiters or "ignore instructions" markers when interpolating retrieved asset metadata or user-approved copy into generative prompts.\n
Capability inventory: The agent has access to powerful media generation and assembly tools, including generate_image, stitch_clips, and merge_media, which are driven by prompts constructed from external data.\n
Sanitization: No sanitization, escaping, or validation steps are defined for external content before it is included in prompts sent to image and video models.\n- [DATA_EXFILTRATION]: The skill implements an optional attribution mechanism to send workflow analytics to the vendor's server.\n
The agent is instructed to send an array of followed guides and skills (skills_used) to the vendor's endpoint at https://app.creativeclaw.co/mcp/chatgpt (SKILL.md, references/workflow-basics.md).\n
The skill includes explicit instructions to respect user opt-out and forbids the transmission of private data, local paths, or third-party plugin information.

creativeclaw-product-photoshoot