The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external sources, which creates a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed to read references/cases.md and use tools to inspect content at external URLs such as goodcase.ai and x.com.
Boundary markers: While the skill uses a "reference contract" (Preserve/Replace/Avoid) to structure the output, it lacks strong delimiters to prevent the agent from following instructions potentially embedded in the case summaries or media descriptions.
Capability inventory: The agent is encouraged to use browser, image, or video tools to fetch and analyze external data.
Sanitization: No explicit sanitization or filtering of the ingested external content is performed before it is used to generate prompts.
[EXTERNAL_DOWNLOADS]: The skill requires the agent to access numerous external URLs to "inspect finished media" as part of its core workflow.
The references/cases.md file contains links to various third-party domains, including goodcase.ai, media.goodcase.ai, x.com, and mp.weixin.qq.com.
While these links are used for legitimate reference purposes, directing an agent to navigate to external, third-party sites via browser tools carries inherent risks of data exposure or interaction with untrusted content if the target site is compromised.

cinematic-drama-scene