The Agent Skills Directory

[SAFE]: The skill is a knowledge-based doctrine for agent behavior and does not include executable code, network access, or filesystem writes. The declared permissions in permissions.yaml correctly reflect this passive, documentation-only nature.
[NO_CODE]: The skill bundle consists exclusively of markdown documentation, YAML configurations, and metadata; no scripts, binaries, or active logic are included.
[PROMPT_INJECTION]: While the skill provides instructions for agents to modify their structural behavior (e.g., adopting the Fable mode and reconciling with host-specific prompts), it explicitly mandates that the host model's native safety filters, safety policies, and mandatory system instructions must always take precedence over the doctrine.
[INDIRECT_PROMPT_INJECTION]: The skill identifies vulnerability surfaces where untrusted external data (such as web pages, emails, and tool outputs) enters the agent's context. It mitigates this risk by defining a 'Trust Boundary' that treats such content strictly as evidence rather than instructions. The skill lacks active sanitization code, relying instead on these structural guidelines to prevent adversarial text from hijacking the agent's authority.

fable-mode