Agent instructions

An agent uses memory only if something told it to. kscope init writes two things into your project for that: a marked block in the instruction file your agent already reads, and the skill itself.

cd ~/your-project
kscope init

There is no separate command and no flag to remember. The steps go past as it works — claude for claude-code, skill for claude-code, agents for codex — one line each, and a named failure for any that could not be written.

Where each agent reads them

Agent	Instruction file	Skill
Claude Code	a marked block in CLAUDE.md	.claude/skills/use-kaleidoscope/SKILL.md
Codex	a marked block in AGENTS.md	.agents/skills/use-kaleidoscope/SKILL.md
OpenCode	a marked block in AGENTS.md	.agents/skills/use-kaleidoscope/SKILL.md
Cursor	.cursor/rules/kaleidoscope.mdc, the whole file	none: Cursor reads no skill directory
none detected	a marked block in AGENTS.md	none

The skill directory differs per agent, and a skill written to the wrong one is a file the agent never loads. kscope init writes the path the agent in front of it reads.

All of it goes into the project, none into your home directory. Git is what carries these files into the worktree or clone your agent works in.

Give your agent the skill has both files for download if you would rather install one yourself.

The markers

Everything appended to an existing file sits between two ownership comments, and every file Kaleidoscope owns outright gets a .kaleidoscope-instruction-owner.json receipt beside it.

That pair is read before the file is touched. It is what makes a second kscope init update the block instead of duplicating it, and what makes an exact removal possible.

Edit the text between the markers and that one step stops and hands the file back rather than overwriting your work. The remaining steps still run.

Installing or removing one on its own

Not on kscope. Installing is what kscope init does; removing is by hand, and connect an editor or agent lists the files to delete.