AI Enablement Cheat Sheet
Category | Technique / Concept | Keywords / Memory Joggers |
Copilot Core | Ghost Text / Autocomplete | Inline suggestions; tab-to-accept; "thinking ahead." |
| Ask / Edit / Agent Modes | Ask: Q&A; Edit: Targeted refactor; Agent: Autonomous tasking. |
| Conversation History | Context memory; ability to "undo" or "checkpoint" to a past state. |
| Copilot Instructions | .github/copilot-instructions.md; global rules for the model. |
| Inline Chat | Cmd+I / Ctrl+I; targeted in-place edit without opening full chat. |
| Slash Commands | /fix; /explain; /tests; /doc — shorthand actions so you don't have to write a full prompt. |
| Commit Message Generation | Copilot drafts a commit message from the staged diff; one click in Source Control. |
| Terminal Assist | AI suggestions for shell commands in the integrated terminal; explains flags and syntax. |
| Chat Participants (@-mentions) | @workspace @github @terminal — built-in Copilot subagents scoped to specific tools or knowledge domains. |
Agentic Logic | Agents vs. Subagents | Primary goal-setter vs. specialized mini-bots (e.g., Security Agent). |
| Plan Mode | AI generates a step-by-step roadmap before writing code. |
| Skills | Custom capabilities or "tools" an agent can call (e.g., "SQL-skill"). |
| Tool-Calling | Allowing AI to run terminal commands, git, or browser-searches. |
| Self-Correction Loop | "Review your own output for errors and provide a fix." |
| Checkpoint / Undo | Restore a prior agent state; rolls back a chain of AI edits to a known-good point. |
| Human-in-the-Loop | Approve or deny each tool call before it runs; prevents unintended file or system changes. |
Context Eng. | MCP (Model Context) | Standard protocol to connect LLMs to local DBs, Slack, or Jira. |
| Adding Context (#) | Pinning specific files, folders, or web docs to a chat session. |
| Context Folding | Compressing long chat histories so the AI doesn't "forget" the start. |
| Prompt Checklists | Giving the AI a list of constraints it must check off before finishing. |
| Context Variables | #file #selection #editor #terminal — explicitly attach specific artifacts to the prompt. |
| Workspace Indexing / #codebase | Semantic search across the entire repo so AI “knows" files it hasn't been shown." |
Prompt Eng. | Few-Shot Prompting | Providing 2–3 examples of the exact input/output you want. |
| Chain-of-Thought (CoT) | "Think step-by-step"; forcing logic before the final answer. |
| Tree-of-Thought (ToT) | Exploring multiple logic paths; backtracking when one fails. |
| Personas | "Act as a Senior Architect" or "Skeptical Peer Reviewer." |
| Multi-Stage Prompting | Breaking a huge ask into 3 prompts (Outline → Draft → Edit). |
| Zero-Shot Prompting | Direct ask with no examples; useful baseline before layering in few-shot. |
| Negative Constraints | “Don't use X"; "avoid class components" — scoping out unwanted patterns explicitly." |
| Output Format Specification | “Respond as JSON"; "use a markdown table" — controlling the exact shape of the response." |
| Iterative Refinement | Follow-up prompts to correct, extend, or narrow a previous response. |
Engineering | SDD (Spec-Driven Dev) | Using a spec.md as the "Source of Truth" for AI implementation. |
| SpecKit / Scaffolding | Automating the creation of folders/boilerplate from a spec. |
| Test-First Generation | Asking AI to write the tests before the functional code. |
| Log-Driven Debugging | Feeding logs/error traces into AI to find the needle in the haystack. |
| Vibe Coding | Interactive, fast-paced, "build-as-you-chat" development style. |
| Code Audit / Security Review | Ask AI to scan for OWASP issues, exposed secrets, or common anti-patterns. |
| Refactoring / Modernization | “Convert to async/await"; "migrate to v2 API" — targeted structural rewrites." |
| Documentation Generation | Auto-generating JSDoc, README sections, or inline comments from existing code. |
| Test Fixing (Red-to-Green) | Feed a failing test + error trace to AI; it diagnoses and patches the failing assertion. |
| Hooks | Creating agent hooks in VS Code to listen in and respond to agent actions and lifecycle. |
Docs & Research | Iterative Drafting | Building documents section-by-section through conversation. |
| Style Injection | "Write this in the style of [Sample A] but the tone of [Sample B]." |
| Perspective Mapping | Asking for "The 3 most common counter-arguments to this plan." |
| Literature Synthesis | Clustering multiple PDFs or links into thematic research groups. |
| Recursive Summarization | Summarizing a summary to find the "core" of a massive doc. |
| Diagram Generation | Mermaid or ASCII diagrams from code descriptions, architecture notes, or data models. |