Skip to content

[init] Claude Code skill /atomic-init wrapping the CLI (v1.1, Approach C) #319

@dep0we

Description

@dep0we

Context

Filed at PR 1 close of the init-wizard arc (#94). Office-hours D2 considered three shapes; Approach C is the "twin distribution" shape originally recommended by Issue #94: ship the Python CLI AND a Claude Code skill that wraps it with chat-driven UX.

Office-hours pivoted to CLI-first (Approach A) for PR 1 because the CLI is the framework canonical primitive and any skill can layer on top later. This issue tracks the layered skill.

What

Create ~/.claude/skills/atomic-init/SKILL.md that:

  1. Detects the operator's intent (chat-driven: "build me an agent that...").
  2. Asks the operator to describe the agent in 2-3 sentences (one Claude exchange).
  3. Calls atomic-agents init <name> --from-template advisor under the hood to scaffold the file tree.
  4. Edits the scaffolded files inline based on the operator's description (Claude reads and adjusts IDENTITY/SOUL/USER/tools/model).
  5. Hands off to atomic-agents doctor for the smoke check + opt-in test call.

Why

Claude Code operators get a chat-driven onboarding shape that compounds with the CLI. Issue #94's original recommendation honored once the CLI surface is stable.

Effort

human ~2 weeks / CC ~4 days. Reuses everything PR 1 + PR 2 build; skill lives in ~/.claude/skills/atomic-init/ outside the framework repo.

Dependencies

Acceptance

  • /atomic-init skill installs under ~/.claude/skills/atomic-init/
  • Calling the skill produces a doctor-clean agent through 1-2 Claude exchanges
  • The CLI it wraps is unchanged (no breaking changes to atomic-agents init)

Deferred from

Office-hours D2 (Approach C, v1.1): #317

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions