modes

skills-anthropic

A marketplace of skills for Anthropic's agent surfaces — Claude Desktop (Chat, Cowork, Code) and the Claude Code CLI.

Skills

Skill	Version	What it does
modes	2.0.0	Persistent response modes via /enterMode, /exitMode, /listModes, /clearModes.
plan2cursor	2.0.0	Copy a .plan.md into ~/.cursor/plans/ and open it in Cursor.

Modes

The modes skill provides six modes. Each mode is a binding contract that overrides defaults and persists across turns until explicitly exited.

Mode	Syntax	What it does
plan	/enterMode plan [dir]	Plan-only writes in [dir] (default ./). Allowed writes are limited to *.plan.md files inside the plan dir; everything else is blocked. Use to produce a Cursor-compatible *.plan.md spec (Cursor's frontmatter format: name, overview, todos, isProject), then exit (or /enterMode agent) to execute. Mutex with agent.
agent	/enterMode agent	Full agency — the default working stance. No write filtering from plan. Mutex with plan.
one-word	/enterMode one-word	Responses are a single word. No punctuation flourishes or trailing clauses.
sbs	/enterMode sbs	Step-by-step. Perform exactly one step, then wait for the user to say "done".
exclude	/enterMode exclude <patterns>	Block writes matching comma-separated gitignore-style glob patterns (e.g. *.log, build/**). Read, Glob, Grep unaffected.
include	/enterMode include <patterns>	Only allow writes matching comma-separated glob patterns. Read, Glob, Grep unaffected.

Invoking the skill — there's one registered slash command, /modes. Pass the action and any args after it, or just use natural language. Both work in every surface.

Action	Slash command	Natural language
Enter a mode	/modes sbs	"enter sbs mode", "turn on sbs"
Enter plan in a dir	/modes plan ./src	"enter plan mode in src/"
Exit a mode	/modes exit sbs	"exit sbs", "turn off sbs"
List active modes	/modes list	"what modes are on?", "show active modes"
Clear all modes	/modes clear	"clear all modes", "reset modes"

Claude Code namespaces plugin skills under /<plugin>:<skill>, so /modes:modes <args> is the explicit form there — /modes <args> works as the shorter alias.

Composability: Multiple modes can be active simultaneously and compose freely, with one exception: plan ↔ agent are mutually exclusive (entering one exits the other). The include and exclude write filters layer on top of plan's *.plan.md allow-list. Output-style modes (one-word, sbs) compose with everything.

Persistence: Active modes persist across turns within a session, and across sessions when an auto-memory directory is available (Cowork). In chat and Claude Code, the directive takes effect for the remainder of the conversation but doesn't survive a new session.

Workflow: plan mode + plan2cursor

The two skills in this marketplace are designed to compose into a plan-then-execute workflow:

1. /enterMode plan [dir] — enter plan mode in the chosen directory. Claude writes only *.plan.md files until you exit.
2. Draft the plan with Claude. The output file follows Cursor's plan format — frontmatter (name, overview, todos, isProject) + body sections.
3. /plan2cursor <path-to-plan-file> — hand the plan off to Cursor. The file lands in ~/.cursor/plans/ (the location Cursor's plans panel watches) and opens in Cursor.
4. Execute the plan in Cursor, or /enterMode agent to execute in place with Claude.

Install

Claude Code (CLI or the Code mode inside Claude Desktop)

Marketplace install — the canonical path:

/plugin marketplace add rushkeldon/skills-anthropic
/plugin install modes@skills-anthropic
/plugin install plan2cursor@skills-anthropic
/reload-plugins

Cowork (Claude Desktop)

Download the .skill file and either drag-and-drop it onto Claude Desktop or use the Save skill button when the file is opened.

• modes/cowork/modes.skill
• plan2cursor/cowork/plan2cursor.skill

Chat (claude.ai)

Upload the per-skill zip via the skills UI on claude.ai.

• modes/chat/modes-chat.zip
• plan2cursor is not shipped to chat — it requires shell + filesystem access (open -a Cursor, write to ~/.cursor/plans/) that the chat surface doesn't provide.

Design

Each skill is packaged per target. The leaves under each skill directory contain a SKILL.md plus a fully-named install artifact tailored to that target's install mechanism: