feat(ai): add style-aware Agent Skills scaffolding to ai init

[Copilot]

Summary

Refines the packaged AI guidance for Plugboard model creation so agents favor YAML-friendly component arguments and decompose processes into components that mirror the real system. Extends plugboard ai init to scaffold AGENTS.md plus Agent Skills-compliant task-specific skills, with a --style option that installs them into the appropriate directory for Agents, GitHub, or Claude workflows.

Changes

• AGENTS guidance

  • Reorganized plugboard/cli/ai/AGENTS.md around a clearer workflow: plan → implement → assemble → export/run.
  • Added explicit guidance to prefer serialisable component arguments so models can be represented in YAML.
  • Added explicit guidance to model real-world entities/stages as separate components rather than collapsing them into monolithic logic.
  • Updated skill references to point to style-specific Agent Skills directories and SKILL.md manifests.

• Packaged skills

  • Added Agent Skills-compliant SKILL.md templates with YAML frontmatter for:
    • creating YAML from a Python-defined process
    • generating diagrams with plugboard process diagram
    • running scenario-specific configs with plugboard process run
    • adding a tune section after eliciting tunable parameters from the user
  • Kept the skill content tool-agnostic and compatible with Claude, Gemini, Codex, and GitHub Copilot conventions.

• CLI scaffolding

  • Updated plugboard ai init to accept --style with:
    • agents → .agents/skills/ (default)
    • github → .github/skills/
    • claude → .claude/skills/
  • Continued to install AGENTS.md alongside the selected skills directory.
  • Added conflict checks so initialization aborts if either target already exists.

• Docs and coverage

  • Updated user-facing docs to describe the new style-specific scaffolded layout and Agent Skills manifest format.
  • Expanded CLI tests to cover:
    • installation of AGENTS.md and .agents/skills/ by default
    • installation into .github/skills/ with --style github
    • failure when AGENTS.md already exists
    • failure when the selected skills directory already exists
    • packaged skill/template presence using SKILL.md

process.dump("model.yaml")
# then use the generated config with:
# plugboard process diagram model.yaml
# plugboard process run model.yaml

[toby-coleman]

@copilot Modify the CLI command so that it takes a --style option, which may be claude, github or agents, and place the skills in one of these directories according to it:
.github/skills/
.agents/skills/
.claude/skills/
Default to agents.
Read https://agentskills.io/specification and make sure the skill files follow the standard, e.g. they should be named SKILL.md, include frontmatter etc.

[Copilot]

@copilot Modify the CLI command so that it takes a --style option, which may be claude, github or agents, and place the skills in one of these directories according to it:
.github/skills/
.ag...

Implemented in 8e1acf8. plugboard ai init now accepts --style (agents default, plus github and claude), installs skills into the corresponding .agents/skills/, .github/skills/, or .claude/skills/ directory, and the packaged manifests are now SKILL.md files with YAML frontmatter.