Agent Manifesto Kit

Agent Manifesto Kit is a curated library of reusable AI instruction capabilities for real projects. It packages skills, agents, pipelines, conventions, and bundles that can be adopted into Claude, Codex, or provider-neutral .ai/ project layouts.

The included agentkit CLI lets you discover the catalog, copy selected capabilities into a project, and optionally ask your preferred AI CLI to adapt the adopted files to local naming, paths, and project conventions.

Install

Install globally:

npm install -g agent-manifesto-kit
agentkit --version

Or run directly with npx:

npx agent-manifesto-kit list

Quick Start

List available capabilities:

agentkit list

Show one catalog view:

agentkit list skills
agentkit list agents
agentkit list bundles

Adopt the Spec-Driven Development bundle for Claude:

agentkit adopt sdd --provider claude

Adopt the QA Automation bundle for browser and automated-test workflows:

agentkit adopt qa-automation --provider claude

Adopt a single skill for Codex:

agentkit adopt brainstorm --provider codex

Adopt into a specific project directory:

agentkit adopt sdd --provider agnostic --dest /path/to/project

What Is Included

Agent Manifesto Kit ships reusable instruction assets:

| Type | Purpose | | --- | --- | | Skills | Focused execution instructions for a specific recurring task | | Agents | Specialized review or analysis roles for delegated judgment | | Pipelines | Sequenced workflows that route multiple capabilities | | Conventions | Shared standards for structure, naming, formatting, or portability | | Bundles | Cohesive groups of capabilities designed to be adopted together |

Bundles install their items into the provider's expected type-specific directories. For example, adopting the sdd or qa-automation bundle for Claude places skills under .claude/skills/, agents under .claude/agents/, and non-capability bundle directories, such as templates, under .claude/<bundle-name>/.

Commands

agentkit list [skills|agents|bundles]
agentkit lint [name]
agentkit adopt <name> [--provider claude|codex|agnostic] [--dest <dir>] [--force] [--cli <cli>]

Command summary:

| Command | Description | | --- | --- | | agentkit list [skills|agents|bundles] | Show the full catalog or one selected view; bundle views include item summaries | | agentkit lint [name] | Check all capabilities, or one named capability, for provider-specific tokens | | agentkit adopt <name> | Copy a capability or bundle into your project |

Global options:

| Option | Description | | --- | --- | | --version, -v | Print the installed version | | --help, -h | Show command help |

Adoption options:

| Option | Default | Description | | --- | --- | --- | | --provider claude|codex|agnostic | claude | Target project layout | | --dest <dir> | Current directory | Project root to receive the files | | --force | Off | Replace existing target files without prompting | | --cli <cli> | None | Run an AI CLI after adoption to adapt files to the project |

Providers

The default adoption step is deterministic. It copies the selected assets and applies only mechanical provider transforms.

| Provider | Destination | Transform | | --- | --- | --- | | claude | .claude/ | Copy Claude-native assets as packaged | | codex | .codex/ | Rewrite .claude/ path tokens to .codex/ and strip Claude-only tools: frontmatter | | agnostic | .ai/ | Rewrite .claude/ path tokens to .ai/ |

Use agentkit lint before or after adoption when you want to inspect capabilities for provider-specific wording.

AI-Assisted Adaptation

Pass --cli to run an AI assistant after files are copied:

agentkit adopt sdd --provider claude --cli claude
agentkit adopt sdd --provider codex --cli codex
agentkit adopt brainstorm --provider codex --cli agy

The assistant receives a structured prompt that explicitly frames the adaptation as an approved, actionable task (not a proposal), asking it to read your project's local instructions and documentation, then adapt the adopted files — including any bundle extras such as SDD templates — to your naming, paths, vocabulary, and conventions.

Codex runs with a writable sandbox and no approval prompts (--sandbox workspace-write --ask-for-approval never) so it can make the edits directly instead of stopping to ask. Claude Code receives the prompt over stdin rather than as a CLI argument, so large prompts never hit OS argument-length limits.

Supported AI CLIs:

claude, codex, agy, aider, opencode, grok, kilo, qwen

Conflict Handling

When a target file already exists and --force is not set, agentkit adopt asks what to do:

Target already exists: .claude/skills/brainstorm
[r] replace  [s] skip  [A] replace all  [S] skip all: _

Use --force in automation when you want existing targets replaced without prompts.

Release History

See CHANGELOG.md for published release history.

License

MIT (c) Alexey Platkovsky