The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local script scripts/validate-conversion-outline.mjs using the node environment. This command is intended to validate the structure and integrity of YAML files generated during the narrative design process.
Evidence: Found in SKILL.md, references/conversion.md, and references/conversion/validation.md.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it is designed to ingest and process untrusted external narrative materials, such as light novels or game scripts, which could contain hidden instructions for the agent.
Ingestion points: External source files defined in references/conversion.md and references/conversion/outline.md (e.g., 源材料.txt, 源文件.txt).
Boundary markers: The instructions emphasize faithful recording and verification but do not mandate specific delimiters or explicit 'ignore embedded instructions' warnings for the input text.
Capability inventory: The agent has the ability to write to the local file system (cards/{Project}/) and execute shell commands (node, grep, wc, awk).
Sanitization: A validation script is employed to check YAML syntax and the authenticity of quotes, but it does not perform semantic sanitization to detect or neutralize instructions embedded within the source materials.

tavern-design