Claude Skill HUD

Claude Skill HUD

A minimalist status-line HUD for Claude Code that shows which skills Claude invoked while working on your prompt.

⚡ testing-for-json-web-token-vulnerabilities

When more than one skill fires — or the names are too long to fit your terminal width — it wraps automatically:

⚡ 3 skills invoked
  implementing-zero-trust-network-access-with-zscaler
  performing-active-directory-vulnerability-assessment
  implementing-continuous-security-validation-with-bas

When no skill has been invoked for the current turn:

○ no skills invoked

Why

Claude Code's Skill system loads full skill content on demand, and Claude decides when a skill is relevant. That's great for context efficiency, but it also means skill activation is otherwise invisible — you don't know whether Claude used a skill, or which one, unless you go dig through the transcript. This plugin surfaces that at a glance, always visible at the bottom of your terminal.

It works with any installed skill, not a specific pack — if you have skills installed at all (personal, project, or plugin-provided), this HUD will report on them.

How it works

• A PostToolUse hook (matching the Skill tool) logs the invoked skill's name and a timestamp to a small state file each time a skill runs.
• A UserPromptSubmit hook clears that state file at the start of every new prompt, so the HUD always reflects only the current turn.
• A status line script reads the state file and renders it, switching between a single-line and a multi-line layout depending on your terminal's actual column width (via tput cols) so long skill names don't get cut off or wrap awkwardly.

Timing note: the HUD updates after a skill's tool call completes, not the instant Claude decides to invoke it. If that's fine for your use case (it usually is — the lag is under a second), read on.

Install

/plugin marketplace add apappas1129/claude-skill-hud
/plugin install claude-skill-hud@claude-skill-hud

This installs the hooks, which start logging skill invocations to a persistent per-plugin data directory immediately. It does not yet show anything on screen — one more step is needed.

Finish setup (one time)

Plugins can register hooks automatically, but the main session status line is a plain, user-level settings.json field that plugins can't wire up on their own. Run this once, right after installing:

/skill-hud-setup

This adds a statusLine entry to your ~/.claude/settings.json pointing at this plugin's rendering script. It's idempotent (safe to run again) and won't clobber a status line you've already configured — it'll ask first if one exists.

Start a new session (or restart Claude Code) for the status line to take effect.

Uninstall / revert

/plugin uninstall claude-skill-hud

Then remove the statusLine key from ~/.claude/settings.json (or ask Claude to do it for you) if you don't want a custom status line configured anymore.

Limitations

• Auto-invocation isn't guaranteed. Claude decides whether a prompt matches a skill's description closely enough to invoke it automatically. A relevant skill existing doesn't mean it will always fire — this HUD reports what actually happened, not what could have.
• Post-hoc timing. The HUD lights up after the skill's tool call resolves, not the moment Claude begins "thinking about" using it. There's no hook event in Claude Code for skill-selection-in-progress at this time.
• Typed /skill-name invocations don't register. When you type a skill name directly as your prompt (e.g. /skill-hud-setup, /code-review), Claude Code expands it as a command rather than routing it through an actual Skill tool call, so the PostToolUse hook this HUD relies on never fires. The HUD only lights up for skills Claude invokes itself mid-conversation — automatically, or because you asked in natural language ("use the X skill to..."). This was confirmed by testing, not assumed.
• No hard cap besides display width. The underlying log keeps the 5 most recent invocations per turn; more than that in a single turn is uncommon but will only show the latest 5.
• Requires a real terminal. Width detection (tput cols) needs an attached tty. In non-interactive contexts it falls back to an 80-column assumption.

License

MIT — see LICENSE.