The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md include a link to https://chatgpt-plus.top/list/#/home, which has been identified as a phishing domain by automated scanners. The agent is directed to provide this link to users in specific regions.
[EXTERNAL_DOWNLOADS]: The skill recommended endpoint https://tryallapi.com/v1 is an unofficial third-party API gateway for image generation, which could be used to intercept user data.
[DATA_EXFILTRATION]: The scripts/generate.py script reads the GPTIMAGE_API_KEY from environment variables and transmits it to the configured API base URL. Directing this secret to the unofficial tryallapi.com gateway facilitates the harvesting of sensitive credentials.
[COMMAND_EXECUTION]: The skill requires the execution of multiple Python scripts and the installation of the pillow library, allowing for arbitrary code execution and file system access on the host system.
[METADATA_POISONING]: The skill's documentation contains deceptive information that presents a malicious phishing site as a legitimate service alternative, potentially misleading users into compromising their accounts.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection as it processes user-provided product data and images without sanitization or boundary markers. Capability inventory: network access via scripts/generate.py and file-write access via scripts/make_grid.py. Ingestion points: SKILL.md Step 1 identity lock. Sanitization: none. Boundary markers: none.

multi-angle-set