What problem does it solve?
This Skill helps teams avoid unintentional high agent autonomy and quality regressions by framing LLM-assisted coding as source-code discipline, then applying explicit quality gates and constrained auto-improve loops to keep changes correct and safe.
Core Features & Use Cases
- Software 3.0 conceptual model: Maps traditional engineering (source code, tests, runtime, debugging) to prompt-as-program workflows so teams can reason consistently about what is being built and validated.
- Autonomy slider (0–5) guidance: Recommends an intentional autonomy level by task type and risk, preventing the default “highest autonomy” failure mode.
- AutoResearch loop constraints: Defines a safe improvement pattern using one editable asset, one scalar metric, and one time box to prevent diffuse edits and metric drift.
- Quality gates for the “vibe hangover”: Provides a practical gate stack (type checking, lint/safety rules, tests, security scanning, review) that compensates for evidence that ungated AI-generated code can be insecure or regress quality.
- Productivity metrics for LLM teams: Replaces lines-of-code metrics with outcome and reliability measures (rework rate, eval pass rate, context-failure rate, time-to-value).
Quick Start
Use the ai-native-development skill when you need to choose an intentional agent autonomy level and define quality gates for an LLM-assisted code change.
