Prompt Engineering Foundations

Developer-first prompt engineering for reliable AI outputs: prompt anatomy, grounding, few-shot design, decomposition, and validation patterns that reduce hallucinations across the software development lifecycle.

Overview

This repository is the foundation layer for working developers who want prompts that are precise, testable, and production-usable. The focus is not on clever wording. The focus is on prompt structure that reduces ambiguity, constrains unsupported claims, and gives models a repeatable path to produce useful results.

    ┌────────────────────────────────────────────────────────┐
    │                Reliable Prompt Anatomy                 │
    │                                                        │
    │  Role -> Context -> Task -> Examples -> Constraints   │
    │                -> Output Format -> Validation          │
    │                                                        │
    └───────────────────────┬────────────────────────────────┘
                            │
                            ▼
          ┌───────────────────────────────────────────┐
          │  Model behavior becomes more constrained  │
          │  - fewer invented facts                   │
          │  - clearer scoping                        │
          │  - more verifiable outputs                │
          │  - easier prompt iteration                │
          └───────────────────────────────────────────┘

What You Will Learn

By the end of this repository, you should be able to:

• structure prompts so the model has less room to invent unsupported details
• separate role, context, task, constraints, and validation into a repeatable workflow
• apply prompt patterns across planning, coding, debugging, review, and documentation tasks
• prepare for the workflow and evaluation repositories that follow this one

Prerequisites

This repository works best if you already:

• use LLMs in developer workflows or want to start doing so
• understand the basics of software delivery, debugging, and documentation work
• want more reliable behavior instead of creative but unstable outputs

Recommended Next Repositories

After this repository, continue with:

• cursor-ai-development-workflows for repo-based AI-assisted development flows
• llm-evals-and-anti-hallucination for measurement and release-gate thinking

Why This Repo Exists

Most weak prompts fail in predictable ways:

1. The role is vague, so the model improvises the standard of quality.
2. The context is incomplete, so the model fills gaps with guesses.
3. The task is underspecified, so the output drifts.
4. The output format is loose, so downstream use becomes fragile.
5. There is no validation clause, so the model never checks itself.

This repository gives you patterns to fix those failure modes systematically.

Core Prompt Framework

Use this as the default prompt skeleton for developer workflows.

Component	Purpose	Hallucination Reduction Effect
Role	Defines the model's operating stance	Reduces behavioral drift
Context	Supplies bounded facts and codebase reality	Reduces unsupported assumptions
Task	States the exact job to do	Reduces scope creep
Examples	Shows what "good" looks like	Reduces formatting and reasoning variance
Output Format	Forces structure	Reduces ambiguous results
Constraints	States what not to do	Reduces fabricated APIs, files, claims
Validation Criteria	Forces a final check	Reduces silent failure

Modules

File	Description
examples/01_prompt_anatomy.md	Breakdown of the base prompt shape with a developer workflow example
examples/02_grounded_code_generation.md	How to ask for code without inviting invented implementation details
examples/03_few_shot_patterns.md	Few-shot prompting patterns for reviews, triage, and structured engineering output
examples/04_prompt_chaining_for_developers.md	How to split complex development tasks into safer prompt stages
templates/base-prompt-template.md	Reusable master template with placeholders
checklists/anti-hallucination-checklist.md	Review checklist for shipping prompts safely

Developer Lifecycle Mapping

SDLC Stage	Prompt Goal	Prompt Behavior
Discovery	Understand codebase reality	Ask for evidence, files, dependencies
Planning	Produce an execution plan	Separate facts from assumptions
Implementation	Generate or edit code	Restrict changes to known files and APIs
Debugging	Diagnose the root cause	Require hypotheses tied to evidence
Testing	Produce verification assets	Require explicit assertions and expected outcomes
Review	Find risks and regressions	Prefer concrete findings over summaries
Documentation	Explain real changes	Ground docs in existing code or diff

Prompting Principles Backed By Market Guidance

The common themes across current vendor guidance are consistent:

1. Be explicit about instructions and response format.
2. Add relevant context instead of assuming the model knows it.
3. Use few-shot examples when consistency matters.
4. Break hard tasks into steps or chained prompts.
5. Evaluate prompts empirically instead of trusting intuition.

For developers, that translates into a practical rule:

If the model could plausibly invent a detail, your prompt should either provide that detail, forbid invention, or define the fallback behavior when the detail is missing.

Best Practices

1. Start every serious prompt with a bounded role and explicit task.
2. Prefer "use only the provided code/context" over broad open-ended prompting.
3. Ask the model to state missing information instead of inferring it.
4. Require assumptions to be labeled as assumptions.
5. Request evidence-backed reasoning for debugging and code review tasks.
6. For code generation, require compatibility with the current file structure and APIs.
7. Add a validation step that checks output against constraints before finalizing.
8. Treat prompt changes like code changes: version them and evaluate them.

References

• OpenAI prompt engineering guidance: roles, reusable prompts, few-shot, context, evals
• Anthropic prompt engineering overview: define success criteria and test empirically
• Google prompt design strategies: constraints, formatting, chaining, self-critique, grounding
• Promptfoo anti-hallucination guidance: evals, rubrics, prompt comparison, regression testing

Author

Dhiraj Singh

Usage Notice

This repository is shared publicly for learning and reference. It is made available for everyone through VAIU Research Lab. For reuse, redistribution, adaptation, or collaboration, contact Dhiraj Singh / VAIU Research Lab.