The Agent Skills Directory

[SAFE]: Systematic analysis of the skill instructions, brand voice requirements, and compliance rubric found no evidence of malicious code, obfuscation, or unauthorized technical operations. The skill adheres to limited, specific workflows.
[INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection because it is designed to process untrusted creative materials provided by partners. Ingestion points: Files, decks, scripts, and emails submitted for review as defined in SKILL.md. Boundary markers: The instructions mandate anchoring each finding to a specific 'filename, slide, page, frame, headline, CTA, footer, timestamp, or visual region,' which restricts the agent's focus to objective observation. Capability inventory: The skill has no access to dangerous tools, shell commands, network operations, or filesystem write permissions; it is limited to generating a text-based report. Sanitization: No specific input sanitization is mentioned, though the rigid output contract further limits the potential impact of adversarial content within the creative assets.

point-partner-creative-review