MathModeling-skills: Instructions file for Codex

Core Philosophy

• The AI owns mechanical correctness; the human owns modeling judgment.
• Start from goals, objects, constraints, data, outputs, variables, relationships, and checkable conclusions.
• Do not start from model names or favorite techniques.
• Separate assumptions, observations, derivations, and validated conclusions.
• Preserve evidence that changes a decision; do not create files merely to prove that a skill ran.

Configuration

planning/session_config.json has two independent controls:

{
  "interaction_mode": "learning",
  "rigor_profile": "lean"
}

• interaction_mode: learning or speed. It changes question density and when AI suggestions are shown.
• rigor_profile: lean or submission. It changes artifact and audit density, never the human-judgment boundary.
• Default to learning + lean in a fresh workspace.
• Use lean while exploring and iterating. Switch to submission only when preparing writer handoff or final assembly.
• For compatibility, read legacy { "mode": "learning" | "speed" } as interaction_mode.

Repository Skill Copies

• .codex/skills/ and .claude/skills/ are two complete, independently usable skill trees.
• Every skill and referenced local resource required at runtime must exist in both trees; neither tree may depend on a wrapper, symlink, or path into the other.
• When a shared skill contract changes, update and validate both copies in the same change.
• Runtime-specific wording may differ only when necessary, but each copy must remain standalone and behaviorally consistent with this policy.
• plugins/mathmodeling-skills/skills/ is the generated distribution copy used by both native plugin manifests. After the two standalone trees agree, refresh it with scripts/sync-plugin.sh and verify it with scripts/sync-plugin.sh --check.
• Keep both plugin manifests and both marketplace catalogs aligned for every release. Bump the version in both plugin manifests and the Claude marketplace entry together.