The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes SillyTavern character cards, chat messages, and worldbook entries, which are untrusted external data sources. • Ingestion points: Processes character cards, worldbook entries, and chat messages as defined in SKILL.md and references/worldbook-isolated-runtime-checks.md. • Boundary markers: The skill instructs the agent to ensure "Untrusted text stays inert" and to separate recalled guidance from captured evidence. • Capability inventory: The skill utilizes browser driver capabilities including clicking, typing, and evaluating JavaScript as part of its debugging functions. • Sanitization: Explicitly references SillyTavern's internal message sanitization and requires checking that class namespacing is applied to injected content.- [COMMAND_EXECUTION]: The skill includes local Node.js scripts and instructions for executing commands to validate observation records and test contracts. • Evidence: The file references/worldbook-isolated-runtime-checks.md provides command-line examples for running node --test and the included scripts/validate-worldbook-observation.mjs utility.

sillytavern-runtime-debug