Skip to content

Piping/agent-kit

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

46 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Kit

Agent Kit is a personal asset manager for prompts, skills, and AGENTS.md files.

The package name is agent-kit. The CLI command is akit.

What It Manages

Agent Kit works with three asset kinds:

  • prompt: a Markdown prompt file
  • skill: a skill package rooted by SKILL.md
  • agents: a project-level AGENTS.md document

The asset store is the source of truth. Installed files are derived copies placed into tool-specific locations when needed.

Asset Store

By default, assets are stored under ~/.akit:

~/.akit/
  assets/
    prompts/<id>.md
    skills/<id>/...
    agents/<id>/AGENTS.md

To use a different location:

export AKIT_HOME=/path/to/akit-home

Installation

From the python/ directory:

pip3 install .

If you prefer running from the repo without installing a script entry point:

PYTHONPATH=. python3 -m agent_kit list

Quick Start

Import a prompt library:

akit add ./agent_kit/raw_prompts --yes

Inspect the imported assets:

akit list
akit show prompt:socratic

Edit a stored asset:

akit edit skill:my-skill

Install a prompt into Codex:

akit install prompt:socratic --target codex

Install a skill package into a project:

akit install skill:my-skill --project /path/to/project

Compare a stored asset with another file:

akit diff prompt:socratic ./some/other/file.md

Push the asset store to the configured personal remote:

akit push

Command Summary

akit list [--kind prompt|skill|agents]
akit show <selector> [--body-only]
akit add <path> [--kind prompt|skill|agents] [--yes]
akit edit <selector>
akit del <selector> [--yes]
akit install <selector> [--target codex|opencode|skill|agents] [--project <path>] [--dest <file>] [--force]
akit diff <selector> <file>
akit diff <selector> --rev <git-revision>
akit push [--remote <name>] [--branch <name>]

Selectors accept either a unique asset id or an explicit kind:id form such as prompt:socratic or skill:mega-skill.

Import Rules

akit add <path> scans Markdown files recursively and classifies them with these rules:

  • SKILL.md defines a skill
  • AGENTS.md defines an agents asset
  • any other Markdown file is treated as a prompt

If a directory contains SKILL.md, that directory is imported as a single skill package. Files inside the package stay bundled with the skill and are not imported separately as prompts.

Stored assets are normalized with internal metadata so they can be listed, selected, and installed consistently. Each stored asset always has:

  • id
  • kind
  • version

When akit add encounters an asset with the same kind:id, it keeps the latest version at the normal asset path and stores version snapshots under a hidden history directory. A new version is created only when the incoming asset does not match any existing version by content fingerprint or source timestamp.

akit edit follows the same versioning rules after the editor exits.

Install Targets

  • prompt + --target codex installs to $CODEX_HOME/prompts/<id>.md or ~/.codex/prompts/<id>.md
  • prompt + --target opencode installs to <project>/.opencode/command/<id>.md
  • skill installs to <project>/.agents/skills/<id>/...
  • agents installs to <project>/AGENTS.md

Installed prompt and skill files use lightweight frontmatter suitable for direct tool consumption. Internal store metadata such as id, kind, and version stays in the asset store and is not written into installed prompt or skill entry files.

Skill installation copies the whole package directory and writes the installed SKILL.md with public-facing frontmatter. AGENTS.md installs as plain Markdown body text.

Version History

The current version stays at the primary asset path:

assets/prompts/<id>.md
assets/skills/<id>/...
assets/agents/<id>/AGENTS.md

Historical snapshots are stored under hidden history paths:

assets/prompts/.history/<id>/v0001/...
assets/skills/.history/<id>/v0001/...
assets/agents/.history/<id>/v0001/...

This keeps selectors and install targets stable while preserving earlier versions for reference.

Editing

akit edit always edits the canonical stored asset inside AKIT_HOME.

  • prompts are opened as files
  • AGENTS.md assets are opened as files
  • skills are opened as package directories

Editor resolution uses:

  1. AKIT_EDITOR
  2. VISUAL
  3. EDITOR
  4. vim

If the edited result is new, Agent Kit writes a new asset version and updates the current version path.

Git-Based Diff

Agent Kit supports two diff modes:

  • compare an asset to another file path
  • compare an asset to a Git revision inside the asset store

Examples:

akit diff prompt:socratic ./drafts/socratic.md
akit diff skill:mega-skill --rev HEAD~1

For revision-based diffs, the asset store directory itself should be a Git repository.

Git Workflow

When AKIT_HOME is a git repository, mutating commands integrate with git automatically:

  • akit add
  • akit edit
  • akit del

For each command:

  • only the paths touched by that command are staged
  • unrelated working tree changes are left alone
  • one conventional-commit-style commit is created when there is an actual change

Examples:

feat(akit): add prompt socratic v1
feat(akit): update skill design-review v4
feat(akit): delete agents work-default

akit push is always manual. It never runs automatically from add, edit, or del.

AKIT_HOME Configuration

Git-related settings are read from AKIT_HOME/config.toml.

Example:

[git]
remote = "origin"
branch = "main"
url = "git@github.com:you/agent-kit.git"

On akit push, Agent Kit ensures the configured remote exists, fetches it, rebases onto the target branch when it already exists remotely, and pushes the current branch.

Typical Workflow

  1. Collect prompts, skills, or AGENTS.md files from different directories.
  2. Import them into the asset store with akit add.
  3. Review what is available with akit list and akit show.
  4. Install only the assets needed for the current tool or project.
  5. Use akit diff to compare stored assets with files or earlier revisions before updating installations.

About

AI agents tool vaults

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages