I build small, honest-by-default tools for AI-assisted development — the kind that tell you the truth about your agent skills and your pull requests instead of rubber-stamping them.
| what it does | |
|---|---|
| skillport | Static cross-agent skill linter — will your SKILL.md work on Claude Code, Codex, and Cursor? Catches what breaks before you publish. |
| skillmatrix | Behavioral cross-agent skill testing — run a skill headlessly on each agent and assert deterministic outcomes (a pass/fail matrix). |
| claimcheck | A CI receipt for the claims your PR makes — extracts factual claims and verifies the checkable ones, honestly flagging the rest as unverified. |
All three: TypeScript, zero runtime dependencies, tested, MIT, and on npm.
Each tool refuses to assert what it didn't check. skillport links every rule to the spec that backs it; skillmatrix only passes a skill that actually produced the right result; claimcheck never confirms a benchmark number it didn't measure.
Honest tooling over confident tooling.