copilot-instructions: Instructions file for GitHub Copilot

Prompt Engineering Standards

Core Principles

Specific context before objective; clear actionable requirements; measurable acceptance criteria; well‑defined output.

Rule: One prompt = one measurable objective.

Prompt Structure (5-Part Framework)

Context (specific file/component) → Objective (what to do) → Requirements (how) → Acceptance (done) → Output (result format).

Context: file src/hooks/useApi.ts lines 10–25
Objective: add exponential retry with jitter
Requirements: keep current interface; configurable timeout; max 3 attempts
Acceptance: existing tests passing; coverage >= 80%
Output: diff of modified function only

---

Prompting Techniques

1. Chain of Thought (CoT)

When: Multi-step logical problems, planning, calculations. Pattern: Request step-by-step reasoning with numbered steps. Risk: Reasoning leakage to users; use "concise reasoning" or "hidden CoT" for production.

Context: Calculate order total with discounts and taxes
Objective: Solve using step-by-step validation
Requirements: Show reasoning in numbered steps; conclude with "Final answer: <value>"
Output: {"steps": [...], "answer": "...", "confidence": 0.0-1.0}

2. Skeleton of Thought (SoT)

When: Long responses with predictable structure (articles, reports, plans). Pattern: Generate skeleton first; expand after approval.

Phase 1 (Outline):

Context: Technical architecture document for microservices migration
Objective: Generate only skeleton with sections and bullets
Requirements: No full text; only titles and bullet points
Output: Markdown outline with ## headers and - bullets

Phase 2 (Expansion):

Context: Approved skeleton from previous response
Objective: Expand each section maintaining bullet structure
Requirements: Keep bullets as subtitles; add details per section
Output: Full markdown document

3. Tree of Thought (ToT)

When: Exploring multiple approaches (design decisions, architecture, strategy). Pattern: Generate alternatives, score paths, choose best.