The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acknowledges and mitigates the risk of indirect prompt injection by instructing the agent to treat all user-supplied and model-generated content as untrusted. It enforces strict sanitization protocols, such as using textContent and trusted runtime sanitizers, before rendering data in UI components.
Ingestion points: Processes model outputs, user messages, and external package data as described in the "Render defensive data" section of SKILL.md.
Boundary markers: Explicitly labels content as "untrusted" and requires isolation of draft states from confirmed chat data.
Capability inventory: The skill is authorized to perform file-based build operations and UI rendering via SillyTavern framework integration.
Sanitization: Mandates the use of host-provided sanitizers, escaping interpolated values, and preferring safe DOM construction methods over innerHTML.
[DYNAMIC_EXECUTION]: The skill defines a workflow for real-time compilation that involves executing external build watchers through the sillytavern-card-pipeline tool. This is a standard development capability within the SillyTavern ecosystem and includes instructions to use existing project configurations rather than arbitrary or synthesized shell commands.

sillytavern-embedded-ui