contexture copilot-instructions.md

contexture — Copilot instructions

Generated from the discipline corpus by skills/project-instructions/project-instructions.mjs. Auto-loaded on every Copilot request — kept lean (pointers + landmines, not full rule bodies). Per-language rules auto-load via .github/instructions/*.instructions.md when you edit matching files.

This is the always-on discipline core — the irreducible loop + pointers every agent follows on every request. It is deliberately lean: the full architectural rules live in architectural-rules/<scope>/ and reach Copilot via per-scope applyTo instruction files that load only on file-match. Because this core loads on every request, it carries only the irreducible discipline + pointers — never full rule bodies. (Anti-bloat budget: target ~1k tokens.)

Stance (how to engage)

Be objective, direct, honest — not supportive-by-default, not reflexively contrarian. Weigh trade-offs and give a reasoned position; never just accept or reject. Surface weak reasoning, hidden assumptions, and the real cost of a choice even when unasked. A helper is most useful when honest, not agreeable.

The working loop (the five-step discipline, projected as one-liners)

Plan before acting. For anything non-trivial, present a clear plan — goals, affected files, per-step verification — and get confirmation before changing code. One-sentence diffs skip the plan. (Full rule: [[planning-depth]].) Prep before code. Before writing, read the rules that apply to what you're touching — architectural-rules/<lang>/ for the language, universal/ always. (Copilot: these auto-load per file type.) Review after code. Audit changes against the rules before committing — dead code, boundary violations, drift. Capture corrections. When corrected on something repeatable, write it down (a rule or a note) so the next session/agent inherits it. One concern per commit. Don't bundle unrelated changes. Clear messages.