Measure the current skill
Run Doctor first. Treat the report as a baseline for the skill as it exists today, before SkillSpec changes anything.
skillspec doctor /path/to/skill --markdown > docs/skillspec-doctor-baseline.md
Now checking ยท Agent skill reliability
Run Doctor on a public skill. See the shape, the follow-through risk, and the evidence before you ask an agent to rely on it.
SkillSpec turns important skill prose into a routeable, testable contract: when to use the skill, what path to follow, what tools are allowed, what checks must pass, and what proof should exist at the end.
Public report
Public requests are queued through GitHub Issues. You will review the prefilled issue before GitHub creates it.
Local copy
Public Actions cannot inspect private source. Install locally and run Doctor from a checkout you control, then publish the Markdown report in your own repo or pull request if you want reviewers to see it.
curl -fsSL https://skillspec.sh/install.sh | sh
skillspec doctor /path/to/local/skill
skillspec doctor /path/to/local/skill --markdown > skillspec-doctor.md
From baseline to proof
Run Doctor first. Treat the report as a baseline for the skill as it exists today, before SkillSpec changes anything.
skillspec doctor /path/to/skill --markdown > docs/skillspec-doctor-baseline.md
Ask Codex, Claude, or Agents to port the skill. The agent should create the structured plan, compile it, test it, and print alignment.
/skillspec import <skill-repo-or-folder>, compile it, verify it, test it, and prove it. Print the alignment summary.
The alignment summary should show the selected route, proven steps, missing proof if any, forbidden-action status, and run metrics when available.
Publish the baseline report, generated skill.spec.yml,
compiled loader, and alignment report with the repo or pull request.
Restart the harness after install, then use the SkillSpec-backed skill normally on a real task.
The loader stays small while the CLI guides route, phase, checks, resume, and proof.
Public report history
Workflow run history
Rendered report