The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts for bootstrapping and auditing. Specifically, run-benchmark.mjs uses execFile to execute the skill's own create-harness.mjs script within a temporary directory as part of a functional self-check. Additionally, the skill generates an init.sh file for the target project which facilitates standard development tasks such as dependency installation and test execution.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for processing untrusted repository data, as it reads package.json and scans file structures to generate project-specific harnesses.
Ingestion points: The detectProject function in scripts/lib/harness-utils.mjs reads the target repository's package.json and file listing to determine the technology stack.
Boundary markers: The generated AGENTS.md and CLAUDE.md templates establish structured startup workflows and 'Definition of Done' criteria to guide agent behavior.
Capability inventory: The skill scripts can create files and templates in a target directory and generate shell scripts for environment verification.
Sanitization: Output generation includes proper escaping, such as escapeForEcho for shell commands and escapeHtml for generated assessment reports.
[DYNAMIC_EXECUTION]: The templates/init.sh script generated by the skill uses node -e logic to dynamically query package.json scripts at runtime. This behavior allows the generated harness to conditionally execute correct verification commands (e.g., npm run lint or yarn test) based on the specific project's configuration without hardcoding the package manager.

harness-creator