name: artifact-compile
description: Strategic planning for AI agent artifacts. Use when you have notes, knowledge, or workflows to turn into reusable agent artifacts (skills, slash commands, AGENTS.md). Helps decide WHAT to build and proposes artifact breakdowns.

Artifact Compile

Strategic orchestrator for turning knowledge into agent artifacts.

When This Triggers

• User has notes or content they want to "compile" into something reusable
• User asks "what should I build from this?"
• User wants to promote notes into agent tooling
• User is deciding between skills, commands, or AGENTS.md

Process

1. Understand the Goal

Ask:
- What are we trying to enable or scale?
- Who will use this? (agent, human, mixed)
- Where will this live? (which agent tools)

2. Analyze the Input

For each piece of content:
1. Summarize the core idea
2. Evaluate:
- Default vs invoked behavior
- Cost of forgetting
- Determinism
- Scope of impact
- Stability

3. Propose Artifact Breakdown

Based on analysis, recommend what becomes:

AGENTS.md content — when:
- Default behavior that should always apply
- High stability
- Violation is usually a bug
- Constrains how work is done

Slash commands — when:
- Invoked intentionally by user
- Workflow or playbook
- Reasoning-heavy
- Context-sensitive

Skills — when:
- Multi-step procedural workflows
- Needs scripts or bundled resources
- Deterministic and repeatable
- Benefits from automatic triggering

Reference only — when:
- Personal or narrative
- Informational
- Low blast radius
- Not reusable

4. Consider Progressive Disclosure

For skills, think about:
- What goes in SKILL.md body (<500 lines)
- What goes in references/ (loaded on demand)
- What goes in scripts/ (executed, not loaded)

Output Format

Recommended Artifacts

Group by type with reasoning:

• AGENTS.md: [proposed snippets with justification]
• Slash commands: [name, purpose, inputs]
• Skills: [name, what it does, what it bundles]

Keep as Reference

Notes that shouldn't be promoted, with reasoning.

Open Questions

Ambiguities requiring human judgment.

References

For detailed specs on each artifact type:
- references/skills.md — Skill anatomy, progressive disclosure, creation process
- references/agents-md.md — AGENTS.md structure and best practices
- references/slash-commands.md — Command syntax, arguments, frontmatter

Load these as needed based on what artifacts are being proposed.

Rules

• Do NOT make decisions unilaterally — this is collaborative
• Do NOT collapse nuance — surface tradeoffs
• Always explain why something should or shouldn't be a particular artifact type
• When in doubt, keep things simpler (reference > command > skill)