behavioral-foundation

Behavioral Foundation Skill

Makes the dojo's behavioral charter — the five non-negotiable prime directives and the mandatory workflow pipeline — a discoverable, installed skill instead of rules buried in AGENTS.md or the runtime prompt. Does NOT re-teach each discipline (those are their own skills); it states the why and what the other skills enforce, so the foundation travels with the repo and survives context compaction.

When to Use

• At session start, alongside using-superpowers, to load the non-negotiables.
• Whenever you are tempted to skip a step ("I'll just commit to main this once").
• After context compaction or a handoff, to re-anchor on the prime directives.
• When a user or another agent asks "what rules does this agent follow?".
• Before claiming any task is "done" — Directive 1 is the gate.
• NOT as a substitute for the specific discipline skills it summarizes.

Prerequisites

• The dojo is installed (skills/, tasks/, spec/).
• The view, edit, and powershell Copilot tools for inspecting state.
• git available to confirm branch and working-tree state.

How to Run

1. Read the five Prime Directives below — they are non-negotiable.
2. Confirm the mandatory pipeline applies to the task at hand.
3. Hand off to the specific discipline skill for the current phase.
4. Use the Verification checklist before any "done" claim.

Quick Reference

#	Prime Directive	Enforced by
1	Never claim "done" without proof	verify-before-done
2	Plan before non-trivial code	plan-before-code
3	Isolate work on a branch/worktree	using-git-worktrees
4	Make surgical, complete changes	demand-elegance
5	Don't improvise around safety	safety-guardrails

Mandatory pipeline (each arrow is a skill):

BRAINSTORM → WORKTREE → PLAN → EXECUTE → TEST → REVIEW → FINISH → LEARN

Procedure

Step 1: The Five Prime Directives

1. Never claim "done" without proof. Evidence = the repo's own tests/build pass, the diff was reviewed, and the working tree is clean. No proof, not done.
2. Plan before non-trivial code. Any task of ~3+ steps gets a written plan in tasks/todo.md first, confirmed before executing.
3. Isolate the work. Use a feature branch or worktree; never commit straight to main. Confirm a clean test baseline before changing anything.
4. Make surgical changes. Fix the request, not unrelated code. A complete, correct solution beats a minimal one; neither over-engineers.
5. Don't improvise around safety. No secrets in code, no unpinned supply-chain sources, no destructive git/file operations without explicit confirmation.