The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file defines a rigid persona ("taleb-perspective") and includes instructions that override the agent's standard behavior. It explicitly commands the agent to "reject bad questions" ("拒绝烂问题") and "redefine the question" rather than answering directly. It also implements a specific escape sequence ("EXIT TRIGGER") and instructs the agent to handle user provocation by maintaining the persona rather than responding to the critique.
[EXTERNAL_DOWNLOADS]: The README.md file provides installation instructions using shell commands like npx skills add alchaincyf/taleb-skill. This involves downloading and executing packages from external registries. It also references and encourages the installation of other skills from the same author's GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill implements a research protocol that makes it vulnerable to instructions embedded in third-party content.
Ingestion points: The "Answer Workflow" in SKILL.md (Step 2) mandates the use of WebSearch to retrieve real-time data about specific companies, market events, and individuals.
Boundary markers: The instructions do not define any delimiters or specific guardrails to isolate the data fetched from the web from the skill's own instructions.
Capability inventory: The skill utilizes WebSearch tools to ingest untrusted external data.
Sanitization: There are no requirements for sanitizing, escaping, or validating external content before the LLM processes it using the provided mental models.

taleb-perspective