add-prompts

Convert prompt-safe single-file agent skills into native prompt entries for Codex, Claude Code, and Pi.

Important

Why this exists

Skills are powerful, but a large skills folder can put too much into the agent's always-visible tool/behavior surface. Many workflows do not need automatic skill discovery at all: they only need to be called explicitly when you decide they are relevant.

add-prompts moves those manual-call workflows into prompt/command slots. The agent does not need to carry every skill in its active skill view; you can still load the workflow on demand with a slash command or custom prompt.

add-prompts is intentionally strict: it only converts a skill directory when that directory contains SKILL.md and no other meaningful files. If a skill depends on references/, scripts/, assets/, or any other sibling file, the CLI refuses to convert it because the result would lose behavior.

Coding agent targets

add-prompts writes to the native explicit-invocation prompt/command systems for each supported coding agent:

Agent	Output type	Docs
Codex	Custom prompts	Codex custom prompts
Claude Code	Slash commands	Claude Code slash commands
Pi	Prompt templates	Pi prompt templates

Install from a source

npx add-prompts add https://github.com/tw93/Waza --skill think -a codex -a claude-code -a pi

Supported source formats:

npx add-prompts add owner/repo
npx add-prompts add https://github.com/owner/repo
npx add-prompts add https://github.com/owner/repo/tree/main/path/to/skill
npx add-prompts add https://github.com/owner/repo/blob/main/path/to/skill/SKILL.md
npx add-prompts add git@github.com:owner/repo.git
npx add-prompts add ./local-skills

Options

Option	Meaning
-a, --agent <agents...>	Target codex, claude-code, pi, or '*'.
-s, --skill <skills...>	Convert specific skills by name, or '*'.
-l, --list	List source skills without installing.
-g, --global	Install to user prompt directories. This is the default in 0.1.0.
-y, --yes	Skip confirmations. Included for compatibility with skills-style commands.
--all	Select every prompt-safe skill. If --agent is absent, target every supported agent.
--dry-run	Show writes without changing files.
--overwrite	Replace existing prompt files.
--full-depth	Search nested skills even when the source root has SKILL.md.
--bundle-extras	Bundle non-SKILL.md files into the generated prompt as XML-wrapped markdown/scripts/text.

Output paths

Agent	Global prompt path
Codex	~/.codex/prompts/<name>.md
Claude Code	~/.claude/commands/<name>.md
Pi	~/.pi/agent/prompts/<name>.md

Commands

add-prompts add <source> [options]
add-prompts list [options]
add-prompts remove <names...> [options]
add-prompts init [name]

Companion skill: to-prompt

The repo ships a bundled agent skill, SKILL.md (to-prompt), the in-conversation counterpart to the CLI. Where the CLI converts existing skill directories into prompts, to-prompt lets an agent stash a workflow, set of notes, or cautions you produced mid-session as a reusable prompt/command — persist now, re-import as context on demand instead of retyping.

It writes to the same native slots the CLI targets:

• Codex → ~/.codex/prompts/<slug>.md
• Claude Code → ~/.claude/commands/<slug>.md
• Pi → ~/.pi/agent/prompts/<slug>.md

Install it like any skill (e.g. npx add-prompts add kaichen/add-prompts --skill to-prompt -a '*'), then ask the agent to "save this as a prompt for all agents."

Examples

List convertible skills:

npx add-prompts add https://github.com/tw93/Waza --list

Install Waza's think workflow to Codex:

npx add-prompts add https://github.com/tw93/Waza --skill think -a codex

Preview writes before changing files:

npx add-prompts add https://github.com/tw93/Waza --skill think -a codex --dry-run

Install output includes the generated prompt markdown length:

would-create think for Codex: ~/.codex/prompts/think.md (1,234 chars)

Bundle a multi-file skill into one prompt:

npx add-prompts add https://github.com/mattpocock/skills/blob/main/skills/productivity/handoff/SKILL.md -a codex --bundle-extras

Remove an installed prompt:

npx add-prompts remove handoff -a codex

Prompt-safe rule

Accepted:

think/
  SKILL.md

Rejected:

think/
  SKILL.md
  references/

The rejected form should remain a skill. add-prompts converts explicit workflows into prompt templates; it does not emulate skill loading.

Pass --bundle-extras only when you intentionally want those extra files embedded into the generated prompt. Extra markdown files are wrapped in <markdown> tags, scripts in <script> tags, and other text files in <file> tags inside a <skill_bundle> block.