The Agent Skills Directory

[SAFE]: No security vulnerabilities or malicious patterns were identified in the skill. The files consist entirely of markdown instructions and YAML configuration for an AI reasoning task.
[PROMPT_INJECTION]: The skill includes instructions to "Set aside the current framing, solution, abstractions, conventions, analogies, and inherited constraints" and "Start over from first principles." While these phrases appear to override previous context, they are standard linguistic markers for the first-principles reasoning technique. They target the problem-solving logic rather than attempting to bypass safety filters or platform guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process a "target" provided by the user to re-derive it from scratch. This represents a data ingestion surface; however, the skill's rigid requirement for decomposing input into "observed facts," "necessary constraints," and "traceable derivations" provides a structured environment that minimizes the impact of embedded instructions in the input data. Furthermore, the skill lacks capabilities (such as network access or file writes) that would allow an injection to perform dangerous actions.
[METADATA_POISONING]: The metadata in SKILL.md and agents/openai.yaml is consistent with the skill's stated purpose and does not contain deceptive instructions or obfuscated content.

first-principles