The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain hidden instructions intended to influence the agent's behavior.
Ingestion points: The agent is instructed to "sample" colors from user-provided websites and to read reference frames (screenshots) from the examples/ directory in SKILL.md.
Boundary markers: The instructions lack explicit delimitation or "ignore embedded instructions" warnings when processing the contents of these external sites or images.
Capability inventory: The skill has the capability to write permanent project instruction files (CLAUDE.md, MOTION.md, brand.md).
Sanitization: There is no sanitization or validation of the data extracted from the external URLs or image frames before it is used to define project-wide behavioral rules.
[PROMPT_INJECTION]: The skill modifies the CLAUDE.md file to inject persistent behavioral overrides.
Evidence: Step 4 of the skill instructs the agent to add a block to CLAUDE.md that mandates the agent must "read MOTION.md in full" and that "Every colour, font, timing and motion value comes from that file" before performing any animation tasks. This creates a persistent constraint on the agent's operations within the project context.

brand-intake