Skilled launches CLI and TUI audits for Claude Code, Codex and Grok Build

• Everlier's launch thread introduced Skilled as a CLI and TUI for auditing installed agent skills, with usage counts, project-level breakdowns, and cross-agent support.
• According to Everlier's TUI overview, the first release already supports Claude Code, Codex, Droid, OpenCode, and Grok Build.
• Everlier's audit post says the audit view flags skills you have not used in the last month and skills that only show up in one project.
• The dashboard in Everlier's early screenshot shows what that sprawl can look like in practice, including 70 skills, 28 projects, and one-off commands that had only been called once.

You can jump straight to the GitHub repo, watch Everlier's launch demo, and inspect Everlier's screenshot of the audit dashboard to see the exact buckets the tool is built around. One useful detail from Everlier's CLI post is that the same data can be queried programmatically or handed back to another agent, not just browsed in a terminal UI.

Skilled is aimed at a new problem that only exists once your coding agents accumulate a real command layer: skill sprawl. Everlier's teaser image made that point visually before launch, with a terminal full of overlapping slash commands.

The core inventory questions come directly from Everlier's launch thread: how many skills are installed, how often they get used, and which projects they belong to. Everlier's TUI overview adds the first concrete support matrix:

• Claude Code
• Codex
• Droid
• OpenCode
• Grok Build

That makes Skilled less like a single-tool add-on and more like a shared audit layer for teams juggling multiple agent shells.

The product looks strongest where it turns messy histories into buckets you can scan fast. Everlier's audit post says the audit pass can surface unused skills from the last month and skills that only one project depends on.

The dashboard shown in Everlier's early screenshot breaks that into concrete lists, including:

• One-off skills, used exactly once
• Single-project skills, tied to only one project
• Summary counts for calls, skills, projects, and sources

In that example, the screen shows 1,087 calls across 70 skills and 28 projects. That is a pretty good argument for why agent tooling now needs its own hygiene tools.

The TUI is only half the pitch. Everlier's CLI post says the same data is available through a CLI, either for scripting or for feeding back into an agent loop.

The implementation details in Everlier's packaging post are unusually specific for a small launch:

• Built with OpenTUI, according to Everlier's packaging post
• Packaged with Bun, per Everlier's packaging post
• Session scanning and caching handled by a standalone Rust binary, also from Everlier's packaging post
• Distributed on npm, PyPI, and via GitHub Releases, as Everlier's packaging post states

That last part matters because it places Skilled in the same cross-stack world as the agents it audits: JavaScript packaging for the app layer, Rust for log processing, and enough distribution options to land outside one ecosystem.