OpenSkills – Universal Skills Loader

Evidence: CLI tool, default project-local install, supports --global and --universal options; has -y to skip prompts; SECURITY.md mentions using HTTPS git clone, cleaning temp dirs, default public repos only; minimal dependencies (commander etc.), no telemetry; has path traversal security tests; has update and remove commands. Deductions: least privilege - tool modifies AGENTS.md and installs skill files but does not clarify required permission scope; user confirmation - has -y skip but does not state default prompting; data flow transparency - does not explain how skill content is loaded and used; sensitive data handling - does not address handling of sensitive data in skills; dependency security - dependencies use ^ ranges, not locked; external effects - modifies filesystem but does not detail impact; rollback - has remove command but does not state rollback to previous state; source attribution - does not explain how skill sources are verified.

Evidence: README and SECURITY.md consistent; tests cover path detection, path traversal, sync logic; dependencies declared in package.json, CI tests; but failure messages - no error handling examples or user prompts. Deductions: self-consistency - docs and tests mostly consistent but not deeply verified; dependency availability - dependencies declared but actual availability not verified; failure messages - no error handling examples.

Evidence: README clearly targets AI coding agent users, supports multiple agents (Claude Code, Cursor, etc.); capability boundaries clear - only handles SKILL.md format; trigger precision - commands and XML markers precise; environment fit - supports Node.js 20.6+, cross-platform. Deductions: audience and scenarios - clear but no specific use cases; capability boundaries - clear but not deep discussion of limitations; trigger precision - commands and XML markers clear but no discussion of false trigger risks; environment fit - supports Node.js 20.6+ but no other environment limitations.

Evidence: README well-structured with quick start, commands, FAQ; install notes detailed, multiple install methods; naming stable, command and option names consistent; examples and FAQ present; known limitations - SECURITY.md mentions some security considerations but no explicit functional limitations; license Apache 2.0; version in package.json but no CHANGELOG; maintenance responsibility - no explicit maintainer or contribution guide. Deductions: information architecture good but lacks detailed docs; install notes detailed but no troubleshooting; naming stable but no version compatibility statement; examples and FAQ present but not all scenarios; known limitations not explicitly listed; license clear but no full text; versioning changelog missing; maintenance responsibility unclear.

Evidence: Output is XML in AGENTS.md and skill files, directly usable by agents; marginal value - provides cross-agent skill loading, unique value; cost-benefit - minimal dependencies, simple install. Deductions: output usability - format clear but actual effectiveness not verified; marginal value - unique feature but no comparison with other tools; cost-benefit - minimal dependencies but maintenance cost not assessed.

Evidence: README claims implementation of Anthropic's Agent Skills spec but no specific link or verification; tests cover some functionality but not independently verified; fact-inference separation - docs distinguish facts and inferences but not explicitly labeled. Deductions: claim traceability - no specific source links; cross-source corroboration - no third-party verification; fact-inference separation - not explicitly distinguished.