ai-native-sdlc

# AI-Native SDLC Take a change from idea to production as a **loop around the agent**, not a linear human relay. Each stage **ends by committing one machine-readable artifact**, and the next stage **begins by reading it**. The chain of commits becomes the audit trail: who asked for what, what the agent produced, and who approved it. ``` Plan -> intent.md Design -> spec.md Build -> plan.md -> diff + tests Test -> a verification target + evals/ Deploy -> PR + REVIEW.md findings -> gated release Maintain -> bands.yaml breach -> writes a new intent.md (loop closes) ``` ## Instructions 1. **Identify the stage.** Read `.sdlc/active` for the current slug; artifacts live in `intent/<slug>/`. If there is no active slug, the work starts at Stage 1. 2. **Read the input artifact before producing the output artifact.** Never write `spec.md` without reading an accepted `intent.md`, and never edit source before `plan.md` is accepted. 3. **Run the gate rather than judging by eye:** ```bash python3 scripts/sdlc_gate.py intent/<slug> <design|build|test|deploy> # exit 0 = gate open, exit 2 = gate closed (reason on stderr) ``` 4. **Write the verification target BEFORE the implementation**, run it, and confirm it FAILS. A test that has only ever been seen green is not evidence. 5. **Implement, then re-run it until green.** If the implementation departs from `plan.md`, update `plan.md` in the same commit. 6. **Commit the artifact at the end ...