The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided SillyTavern character cards (JSON/PNG) which may contain instructions targeting the AI.
Ingestion points: Ingests JSON and PNG card files into a local snapshot directory as defined in references/safe-disassembly.md.
Boundary markers: Mandates the use of a fresh sandbox environment and explicit output roots to isolate processing.
Capability inventory: Performs file read/write, SHA-256 hashing, and extraction of components like regex and helper scripts.
Sanitization: Validates UTF-8 encoding, PNG signatures, and performs semantic round-trip comparisons.
[DYNAMIC_EXECUTION]: The skill identifies and extracts embedded executable components from card data, such as helper scripts and regex patterns.
Evidence: SKILL.md and references/component-model.md detail the extraction and modularization of worldbook entries, regex, and helper scripts.
[EXTERNAL_DOWNLOADS]: The skill's component model allows for the declaration and management of remote runtime dependencies and loaders.
Evidence: references/component-model.md includes a remote_runtime class for assets sourced from Git repositories or CDNs.
[COMMAND_EXECUTION]: The instructions involve invoking extraction tools and build processes to manage the component library.
Evidence: SKILL.md and references/safe-disassembly.md describe the use of extraction adapters and tools to process card data and generate artifacts.

sillytavern-card-components