The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes regex definitions and test fixtures from external JSON files provided as arguments to the scripts. If these files are sourced from untrusted community members or character cards, they could contain instructions designed to manipulate the agent when it reads the trace or validation output. The agent's instructions to "Trace the stages" and "Report" the results create a surface where this injected content could be interpreted by the LLM.\n
Ingestion points: scripts/run-regex-fixtures.mjs and scripts/validate-tavern-regex.mjs read external regex.json and fixtures.json files using fs.readFileSync.\n
Boundary markers: The skill does not implement specific delimiters or safety warnings (e.g., "ignore embedded instructions") when presenting the output of replacements to the agent.\n
Capability inventory: The skill possesses the ability to read local files and perform JavaScript string replacements based on dynamic regex patterns.\n
Sanitization: While the scripts validate the JSON structure and regex syntax, they do not sanitize the content of the input or replaceString fields for potential prompt injection patterns before the agent processes them.\n- [DYNAMIC_EXECUTION]: The parseRegex function in scripts/validate-tavern-regex.mjs uses the new RegExp() constructor to compile search patterns provided in the input JSON. While this is the intended functionality for a regex validation tool, it involves the dynamic compilation of logic defined in external data.

sillytavern-render-regex-pipeline