The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze user-provided arguments, which presents a surface for indirect prompt injection.
Ingestion points: User-supplied text provided in response to coaching prompts in SKILL.md.
Boundary markers: The skill employs [WAIT] markers to strictly control the flow of interaction, preventing the model from processing large blocks of potentially malicious input at once.
Capability inventory: There are no dangerous tools enabled; the skill lacks file system access, network transmission capabilities, or shell execution.
Sanitization: The instructions require the agent to normalize input by rewriting arguments into a clean 'premises → conclusion' format before analysis, which helps neutralize embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill documentation contains links to external websites for educational purposes.
Evidence: References to deciqai.com, github.com/deciqai, stanford.edu, mit.edu, and doi.org are found in SKILL.md and references/sources.md.
Context: These links point to official vendor documentation, GitHub repositories, and peer-reviewed academic papers supporting the skill's logic methodology.
[SAFE]: No obfuscation, persistent mechanisms, or credential harvesting patterns were detected. The skill's behavior matches its stated educational purpose.

logical-fallacies