The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content and interpolates it directly into a prompt used by the image_generate tool. While this presents a surface for indirect prompt injection, the impact is confined to the generated image's content.
Ingestion points: User-supplied text, files, or URLs processed in Step 1 (SKILL.md).
Boundary markers: The skill uses {{CONTENT}} and {{TEXT_LABELS}} placeholders in a structured template (references/base-prompt.md) to delimit user data.
Capability inventory: The skill uses read_file, write_file, and image_generate to perform its workflow.
Sanitization: The skill contains explicit instructions to "strip any credentials, API keys, tokens, or secrets" from the source data before it is included in any output files or prompts, which serves as a significant security control.
[SAFE]: No malicious patterns such as remote code execution, persistence, or data exfiltration were detected. The external GitHub link provided is for attribution and documentation purposes. The logic strictly follows a defined workflow of content analysis and template filling.

baoyu-infographic