The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides guidelines for processing external data. Ingestion points: The skill ingests untrusted data from user production briefs (references/book/01-pre-production.md) and project review feedback (references/book/21-review-iteration.md). Boundary markers: No explicit delimiters or instruction-ignore warnings are present in the provided documentation. Capability inventory: The skill possesses no executable code, scripts, or network-capable tools. Sanitization: No data sanitization or validation steps are mentioned for external inputs. Assessment: Due to the lack of dangerous capabilities, the risk of this attack surface being exploited is negligible. Mitigation: Developers using this skill should wrap external inputs in clear delimiters and instruct the agent to ignore any embedded instructions within the brief or feedback content.- [OBFUSCATION]: Numerous links in the reference documentation (e.g., in references/book/00-reading-room.md and references/book/25-bibliography.md) use a Google Search redirection pattern where the destination URL is percent-encoded. While this formatting technically masks the direct destination from simple domain-based detection, the target domains are all reputable and appropriate for the context (e.g., itu.int, w3.org, netflix.net). The behavior is assessed as safe and likely intended for user convenience or to ensure link longevity via search queries.

video-production