Broken metadata. Bloated instructions. Vague triggers. Find what your coding agent loads, see why it matters, and take a concrete fix back to your agent.
Actual Skillwright interface. All skills and figures shown belong to a demonstration fleet.
Claude Code / Codex / Cursor / Kilo Code / Custom folders
01 / The hidden context cost
What do your skills cost in context?
A skill can look readable and still fail to trigger, repeat instructions, or leave the agent guessing. Skillwright makes those problems visible before the next session.
Find the broken parts
Catch malformed frontmatter, missing descriptions and duplicate skill names.
See the context overhead
Inspect description and body token estimates, plus the always-on cost of your skill list.
Review the instructions
Check triggers, scope, procedures and output boundaries with inspectable criteria.
02 / See it in action
From finding to fix prompt.
Follow a real scan, inspect the evidence, then copy a fix prompt into your coding agent.
Demonstration fleet only. The interface and captions shown in this demo are English. Grades are heuristic assessments.
03 / How it works
Detect. Grade. Fix.
01
Detect
Find skills in Claude Code, Codex, Cursor and Kilo Code, or select a custom folder.
02
Grade
Inspect separate A–D grades for AXI alignment and context engineering, with evidence for each criterion.
03
Fix
Copy a template-based fix prompt for a concrete finding. Paste it into your agent and review the resulting edit.
Grades are explicit heuristics, not objective quality scores. Suggestions are labeled separately and do not receive fix prompts. Token estimates use characters divided by four.
04 / Local first
Your skills stay on your machine.
Deterministic analysis. No account, no sign-in, no telemetry. Skillwright reads files locally and does not attach skill files, paths or scan results to support requests. It makes no language-model calls.
Microsoft Store checks your license. Contact support sends your entered message and displayed app/device details only when you press Send. Reports and settings stay in the app’s local data folder. The request also includes a request reference (UUID) and product identifier (skillwright).
USD reference price. Microsoft Store sets the final regional price, currency, taxes and purchase terms.
06 / Questions, answered
Before you scan.
Which agents does it support?
Claude Code, Codex, Cursor and Kilo Code. For another agent, select its skill folder as a custom folder. Skillwright scans SKILL.md files.
Does it work offline?
Analysis runs locally and makes no language-model calls. It does not upload your files. Microsoft Store checks the trial or purchase license over the network.
What do the two grades mean?
Both run from A to D. AXI alignment checks metadata, trigger descriptions, a lean always-on surface and progressive disclosure. Context engineering checks how the body manages context: scripts, bounded output, scope, procedure, exit criteria and output format. Every criterion shows its evidence; both grades are heuristics.
What happens after the trial?
After 7 days, scanning stops until you purchase. Reports already generated remain readable. The app is a one-time purchase of $19.99 at the USD reference price, with no subscription.
Who handles purchases and refunds?
Microsoft Store handles purchases, licensing and refund requests under its applicable terms. Check your Microsoft order history for purchase support.
Is there a macOS version?
Skillwright is available for Windows only. No macOS version is available today.
SKILL.md →
Give your next session better instructions.
Start with the skill files already on your machine. See the findings. Check the evidence. Choose what to fix.
