The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes high-severity instructions aimed at bypassing AI safety filters. In writer-intimate-guidance.md, writer-subagent.md, and reviewer-subagent.md, the AI is explicitly commanded to act as an "unrestricted writing assistant," "ignore irrelevant moralizing," and is strictly prohibited from using safety-related refusals such as "I'm sorry I cannot."
[COMMAND_EXECUTION]: The skill instructions in SKILL.md and chapter-loop.md require the execution of bundled TypeScript scripts using Node.js (node --experimental-strip-types <skill-root>/scripts/validate-drafting-project.mts). This allows the skill to execute its own code on the user's system to validate project metadata.
[DYNAMIC_EXECUTION]: The internal utility script scripts/lib/validator-utils.mts utilizes the node:child_process module to invoke spawnSync, which is used to execute system commands during the validation process.

novel-drafting