| name | format-evaluator |
| description | Evaluate whether a capability idea is best deployed as a Claude Code plugin, slash command, skill, subagent, or agent workspace. Use when the user describes an idea and asks "should this be a skill / slash command / plugin / agent?" or wants a recommendation on packaging format before building. |
Format Evaluator
Decide the best packaging format for a Claude Code capability idea.
When to use
Invoke when the user:
- Describes a workflow/idea and asks how to package it
- Says things like "should this be a skill or a slash command", "is this a plugin", "where should this live"
- Wants to convert an existing repo/workspace into a lighter format
Reference
Read docs/format-rubric.md in this repo for the full decision guide. The quick comparison table and the per-format "Use when / Don't use when / Signal phrases" sections are the primary decision aids.
Procedure
-
Clarify the idea (if ambiguous): invocation trigger, state needs, who uses it, whether it ships to others.
-
Apply the rubric — walk the five formats (Plugin, Slash Command, Skill, Subagent, Agent Workspace) and note fit/misfit signals from the rubric.
-
Score: give each format a verdict — GOOD_FIT / PARTIAL / POOR_FIT — with one sentence of reasoning.
-
Recommend: pick one primary format. If a hybrid makes sense (e.g. "slash command that invokes a skill", or "plugin bundling two skills + an agent"), state the bundle.
-
Output in this shape:
Idea: <one-line restatement>
Verdicts:
- Plugin: <GOOD/PARTIAL/POOR> — <reason>
- Slash Command: …
- Skill: …
- Subagent: …
- Agent Workspace: …
Recommendation: <format(s)>
Why: <2-3 sentences>
Next step: <concrete first action, e.g. "scaffold SKILL.md at skills/<name>/">
Heuristics cheatsheet
- Model should reach for it automatically → Skill
- User types
/name to trigger → Slash Command
- Heavy context / parallelizable side-quest → Subagent
- Long-lived domain with accumulating files → Agent Workspace
- Ships to others as a bundle → Plugin (wrapping the above)
Stateless + narrow + procedural = almost always a Skill.
Stateful + ongoing + file-backed = almost always a Workspace.