Slash Command

/skill-validate

From skillit

Score a Claude Code skill's token efficiency and structure, returning a graded scorecard with prioritized fixes.

Popularity

Stars

Invocation

How this command is triggered — by the user, by Claude, or both

Slash command

/skillit:skill-validate path-to-skill-dir-or-SKILL.md

Model invocable

No pre-commands

Tool Access

This command is limited to the following tools:

TaskRead

Context Preview

The summary Claude sees in its command listing — used to decide when to auto-load this command

# /skillit:skill-validate

Score a skill against the `skillit` rubric and report a graded scorecard. The heavy reading and scoring runs
in a subagent so it stays off the main context — itself a token-efficiency demonstration.

## Steps

1. **Resolve the target.** Use `$ARGUMENTS` as the skill path if given; otherwise default to a `SKILL.md`
   in the current working directory. If none is found, ask the user for the path.

2. **Delegate to the scorer.** Launch the `skill-scorer` agent (via the Task tool) with the resolved path.
   The agent reads the rubric, the target `SKILL.md`, and its `r...

Command Content

29 lines · ~365 tokens

Stats

Stars1

MaintenanceExcellent

Last CommitJun 16, 2026

Actions

View Source View Plugin View on GitHub View README

Stats

Actions

/skillit:skill-validate

Score a skill against the skillit rubric and report a graded scorecard. The heavy reading and scoring runs in a subagent so it stays off the main context — itself a token-efficiency demonstration.

Steps

Resolve the target. Use $ARGUMENTS as the skill path if given; otherwise default to a SKILL.md in the current working directory. If none is found, ask the user for the path.
Delegate to the scorer. Launch the skill-scorer agent (via the Task tool) with the resolved path. The agent reads the rubric, the target SKILL.md, and its references/, then returns a structured scorecard. Do not score inline — let the subagent do it so the main context stays clean.
Present the scorecard exactly as returned: the per-dimension table, the measured facts (description length, body size, allowed-tools), the total grade (X/12) and band, and the prioritized fix list.
Prompt the measurement loop. Remind the user to run /context (compare to their baseline) and /cost to see the real footprint and spend — the rubric scores structure, these commands measure the actual tokens. 🍳

Address the highest-leverage fixes first: description before body before runtime.