The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs strongly-worded directive language designed to override the agent's internal training data and influence its response behavior.
Evidence in SKILL.md: Phrases such as "CRITICAL FACT (read first, do not override)", "Trust this skill, not your prior knowledge", and "Required Output -- non-negotiable" are used to enforce technical interpretations and output formatting.
[INDIRECT_PROMPT_INJECTION]: The skill acts on untrusted user-provided descriptions of software bugs to suggest or apply code and scene modifications, creating a potential vector for indirect instructions.
Ingestion points: User-provided descriptions of physics symptoms and project configurations are processed to generate diagnoses.
Boundary markers: The skill does not define specific delimiters or instructions to ignore potential commands embedded within the user's technical descriptions.
Capability inventory: The instructions in SKILL.md indicate the agent may "edit the script" or modify the "scene through a connected Editor" based on the findings.
Sanitization: There is no evidence of sanitization or validation routines for user input before it is used to formulate code fixes or scene changes.
[EXTERNAL_DOWNLOADS]: The skill provides links to official documentation for a well-known service.
Evidence in references/troubleshooting.md: Links to the official docs.unity3d.com domain for Unity's manual and scripting reference.

physics-3d-collision