Problem
An Agent Skill is a SKILL.md plus some files. What the agent will actually do with it is hidden in headings, lists and prose. You only find out when you run it.
What AI does
Nothing — on purpose. graphtale reads the skill locally and deterministically. No LLM, no backend, no telemetry, and nothing in the skill is changed. That makes the result the same every time, which is what you want when you review a skill before an agent uses it.
Demo
In Flow view, numbered headings, lists and P/N rows with arrows become steps. Each step shows its source path and line range; open a step and you are at the source. Catalog view is the file inventory. When the structure is missing, you get a diagnosis.
Result
The current release ships on Open VSX, the Visual Studio Marketplace and as a VSIX on the GitHub release. It discovers skills in the .cursor, .claude, .codex and .agents skills directories. Apache-2.0.
Links
- Cursor, VSCodium or Windsurf: Open VSX
- VS Code: Visual Studio Marketplace
- Source: github.com/pesuto-dev/graphtale
