Core skill and personal wrapper

A proposed Claude Code interface that accepts caller paths and keeps personal policy outside the core.

This page specifies a future Claude Code integration with the implemented app. Neither /diagram-gen nor /my-diagram-gen is installed by this repository or its initializer. The options below are a proposed skill contract, distinct from the implemented application CLI. Running these examples as shell commands will not work.

Core responsibilities

The core skill should read project context, establish the diagram brief, initialize or resume a session, choose useful tone references, generate and validate candidates, present the review, and carry feedback into later rounds.

The core accepts the caller's paths. It does not choose a user's personal logs directory, require a particular home-directory layout, or install itself as an incidental part of initializing a session.

Proposed skill options

Option	Meaning
`--out <directory>`	Destination for a new session. Required when creating a session unless the caller provides it through another explicit interface.
`--project <path-or-reference>`	Project context to inspect for behavior, terminology, and design rules.
`--session <directory>`	Resume an existing session and its recorded rounds.
`--count <number>`	Requested number of candidates for the next exploration.
`--tones <id,id,...>`	Optional references the user explicitly wants considered.

Use --out for creation and --session for continuation; they should be mutually exclusive in a single invocation. The skill can infer project context from the caller's working project when it is unambiguous, but the output destination remains an explicit input to the core contract.

Example future invocations:

/diagram-gen --project . --out ../feature-diagrams --count 10
Explain this feature for its help dialog and show useful visual directions.

/diagram-gen --session ../feature-diagrams
Continue from r01-c07. Keep its composition and labels; simplify the arrows.

--tones selects references for generation. It does not guarantee that copying a reference unchanged explains the feature, and it does not turn the CLI into an SVG rendering template language.

Personal wrapper responsibilities

A wrapper such as /my-diagram-gen can apply personal conventions before calling the core:

Resolve a preferred output directory using the user's existing helper or settings.
Create a unique session destination within that directory.
Supply --out and the current project reference to the core.
Apply preferred host/port discovery and presentation conventions when starting the viewer.
Supply a preferred exploration count only when the current request has not specified one.

The wrapper passes resolved values; it does not add its filesystem policy to the engine, initializer, or shared core skill.

Conceptually:

/my-diagram-gen
  resolve this user's output convention
  call /diagram-gen with an explicit destination and project context

The core remains reusable by other people and agents that choose a different working directory.

Existing app interface

The future skill can already operate against these implemented boundaries:

Initializer: create a small session project at an explicit destination.
Files: write the brief, round metadata, candidate metadata, and SVG assets.
Catalog commands: inspect available tone references.
Validation: check session structure and exportability.
zfb commands: run the local viewer and produce a static site.
Review transfer: read copied feedback or a review JSON file supplied by the user.
Export: retrieve the exact chosen SVG for integration.

An installable skill remains future work. Its core instructions should stay short, with detailed schema, authoring, and review guidance in references that can be read when needed.