Evalkit

Plugin evalkit from marketplace tashtit in tashtit/marketplace. Commands read from the repo's manifest — yesterday.

Skills (7)

• benchmark-skills

  Usage: /benchmark-skills <skillA> <skillB> <task…> — requires two skills and a task. Compare two specific skills head-to-head by running the same coding task under each, in isolated git worktrees, then comparing gates, cost, tokens, and review findings. Expensive and side-effecting; spawns two headl

• compare-models

  Usage: /compare-models <task…> — requires a task; the two models are picked interactively afterward. Compare two models on the same coding task by running each in an isolated git worktree and diffing gates, cost, tokens, and review findings. Expensive and side-effecting; spawns two headless agent se

• compare-skills

  Usage: /compare-skills <skillA> <skillB> — requires two skills; static, no task or worktrees. Compare two skills' authoring quality side-by-side by reading their files and synthesizing a difference summary. Static analysis only; runs no coding task and creates no worktrees. Use when asked which of t

• evalkit

  Evaluate and compare skills and models. Route a request to the right evalkit tool — static authoring review of one or two skills, dynamic worktree harnesses that run the same coding task while varying a skill's presence / one skill vs another / one model vs another, or cleanup of the worktrees those

• evaluate-skill

  Usage: /evaluate-skill <skill> <task…> — requires a skill and a task. Measure whether a specific skill changes coding-task outcomes by running the same task twice — once with the skill present, once with it deleted — in isolated git worktrees, then comparing gates, cost, tokens, and review findings.

• remove-worktrees

  Usage: /remove-worktrees [list | all | --prefix <p> | <name>…] [--keep-branches] — no args lists only; nothing is removed without an explicit target. List and remove git worktrees and their associated branches, with a confirmation gate for any worktree holding uncommitted or unpushed work. Defaults

• review-skill

  Usage: /review-skill <skill> — requires one skill; static, no task or worktrees. Review a single skill's authoring quality — structure, description and triggering effectiveness, and progressive disclosure — by reading its files. Static analysis only; runs no coding task and creates no worktrees. Use

Subagents (1)

• skill-reviewer

  Read-only reviewer of a single skill's authoring quality. Use when the user has created or modified a skill and asks to "review my skill", "check skill quality", "improve skill description", or wants to ensure a skill follows best practices. Reviews structure, description/triggering effectiveness, c

Similar plugins