The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands and local scripts as part of its mandatory workflow.
Evidence: Execution of pnpm index-rules in SKILL.md to register new rules.
Evidence: Execution of node .claude/tools/cli/validate-integration.cjs <rule-path> for artifact validation.
Evidence: Shell command test -f .claude/rules/<rule-name>.md used to check for existing files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) because it creates persistent instruction files (rules) from user-provided input.
Ingestion points: args.content and args.name in SKILL.md and scripts/main.cjs.
Boundary markers: None. The rule content is written directly to markdown files without encapsulation or 'ignore' instructions.
Capability inventory: The skill has the capability to write files to the local file system (.claude/rules/) and trigger indexing commands.
Sanitization: scripts/main.cjs includes a sanitization step for the rule name (replace(/[^a-z0-9-]/g, '-')), which effectively prevents path traversal attacks. Content is checked for being non-empty.
[REMOTE_CODE_EXECUTION]: The skill dynamically loads local Node.js modules that are external to the skill bundle but part of the expected host environment.
Evidence: require('.claude/lib/creators/creator-commons.cjs') and require('.claude/lib/creators/companion-check.cjs') in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill requires research using well-known external services for best practices.
Evidence: Instructions to use the Exa tool and arXiv via WebFetch for implementation research in the 'Research Gate' section of SKILL.md.

rule-creator