The Agent Skills Directory

[OBFUSCATION]: The file references/cases.md contains hidden Unicode characters used for steganographic prompt injection. This technique embeds data between visible characters that a human reviewer cannot see, but which the AI agent processes as instructions, potentially leading to unauthorized or malicious behavior.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow requires the agent to ingest and analyze data from external URLs (specifically goodcase.ai and x.com) using image, video, or browser tools to "inspect finished media." This ingestion of untrusted external content without explicit sanitization or boundary markers creates a vulnerability to indirect prompt injection.
Ingestion points: Remote links in references/cases.md (e.g., media.goodcase.ai, x.com) are accessed via agent tools.
Boundary markers: The skill provides no instructions to treat external data as untrusted or to ignore embedded instructions.
Capability inventory: The agent uses browser and media tools to process these external resources.
Sanitization: There is no evidence of filtering or sanitization of the remote content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching media and webpage content from external domains during its execution. While these sites (like X/Twitter) are well-known, the reliance on third-party web content introduces potential risks if those sites or the specific links are compromised, especially when combined with the instructions to treat that content as primary evidence for artifact generation.

sports-extreme-stunts