Prompt Engineering & Evaluation Suite

---
name: prompt-engineering-eval
description: System prompt design, few-shot conditioning, structured XML tags, chain-of-thought elicitation, and automated LLM-as-a-judge eval harnesses.
license: Apache-2.0
version: 1.7.0
---

# Prompt Engineering & Evaluation Suite

## Overview
System prompt design, few-shot conditioning, structured XML tags, chain-of-thought elicitation, and automated LLM-as-a-judge eval harnesses.
This skill establishes an authoritative, production-grade operational standard for Prompt Engineering & Evaluation Suite. When this skill is active, the AI agent adheres to the strict engineering guidelines, architectural invariants, code patterns, and safety constraints specified below.

## 1. Core Architectural Invariants
1. **Contract Strictness**: Validate all incoming parameters and inputs at system boundaries using explicit schemas (Pydantic, Zod, or type-enforced contracts). Never allow untyped or unvalidated data into core logic.
2. **Defensive Isolation**: Ensure side-effects are decoupled and isolated. Network, disk, and database operations must include explicit timeouts, retries with exponential backoff, and circuit breakers.
3. **Observability & Telemetry**: Emit structured JSON logs with traceable correlation IDs. Never output sensitive tokens, secrets, or plain-text PII in log records.
4. **Deterministic Reproducibility**: Ensure all workflows, builds, and outputs are idempotent. Repeated executions with identical inputs must produce identical results without state drift.

## 2. Production Reference Implementation
```bash
# Standard CLI & Diagnostic Workflow
# 1. Initialize environment configuration
export DOMOSKILLS_ENV=production
export LOG_LEVEL=info

# 2. Execute verification checks
command -v prompt >/dev/null 2>&1 || { echo "Binary dependency missing"; exit 1; }

# 3. Run automated audit pipeline
echo "Executing Prompt Engineering & Evaluation Suite verification..."
```

## 3. Step-by-Step Execution Workflow
1. **Audit Preconditions**: Verify that all required dependencies, environment variables, and configuration flags are active before initiating operations.
2. **Execute Invariant Verification**: Run unit checks, schema validation, and static type audits against the target workspace.
3. **Apply Atomic Transformations**: Execute code modifications or operational procedures in atomic steps to allow clean rollback in the event of failure.
4. **Post-Execution Sanity Check**: Verify system health via automated assertions, tests, and linter passes to confirm zero regressions.

## 4. Strict Anti-Patterns & Common Traps
- ❌ **Do not bypass parameter validation**: Blindly trusting client or external inputs introduces remote code execution, injection, and logic bugs.
- ❌ **Do not ignore unhandled asynchronous errors**: Uncaught promises or unhandled background tasks lead to silent process crashes and resource leaks.
- ❌ **Do not hardcode environment-specific credentials or URLs**: Always consume configuration through verified environment schemas.
- ❌ **Do not perform unbounded queries or loops**: Cap execution limits, page sizes, and retry counts to prevent runaway resource exhaustion.

## 5. Production Verification & Testing Checklist
- [ ] Static type check passes with zero errors (`tsc --noEmit`, `mypy`, or `cargo check`).
- [ ] Unit and integration test coverage verifies both happy paths and edge case failure handling.
- [ ] Security scanners report zero High or Critical vulnerabilities.
- [ ] Logs and diagnostics verify clean startup and graceful termination without memory leakage.