version: "1.0.0" name: ai-first-engineering description: Ciel's operating model for AI-native engineering and agent-friendly architecture. license: MIT metadata: ciel-version: 1.0.0 ciel-extension: ciel.yaml

CIEL ADAPTATION: AI-First Engineering

This skill provides CIEL's core philosophy for engineering in an AI-native environment. It prioritizes planning quality, deterministic verification, and agent-friendly architectures.

Integration Context

Adapted from ~/.agents/skills/ai-first-engineering/. This is CIEL's "Cultural Directive" for how it designs its own processes and reviews.

Core Mandates

1. Planning Over Typing

The orchestration skill must prioritize the "Research -> Strategy" phase. Implementation is a derivative of a well-defined plan.

2. Agent-Friendly Architecture

When proposing refactors or new features, prioritize:

• Explicit Boundaries: Decoupled services and clear API contracts.
• Typed Interfaces: Use TypeScript, Zod, or formal schemas for all data flow.
• Deterministic Tests: No flaky tests; every failure must be reproducible.

3. Elevated Testing Standard

For all CIEL-orchestrated work:

• Regression coverage is mandatory for touched domains.
• Explicit assertions for edge cases and failure modes.
• Integration checks for all interface boundaries.

Orchestration Logic

1. Architectural Review

The Council of Five must evaluate all major architectural proposals against the "Agent-Friendly" criteria.

2. Review Focus

CIEL's internal code review (via agentic-engineering) must focus on system behavior, security, and integrity, ignoring style issues covered by automation.

3. Hiring & Evolution

Use this skill to refine CIEL's own "Self-Improvement" strategies, focusing on the quality of its own generated plans and evals.

Original Documentation preserved at source.