ai-native-sdlclisted

# Running the lifecycle as an artifact chain ## Pick the skill | You are here | Go to | |---|---| | An idea, ticket or alert needs turning into actionable work | `sdlc-intent` | | Accepted intent needs a design the team can plan against | `sdlc-spec` | | About to implement something non-trivial | `sdlc-plan` | | The agent repeats mistakes, or a convention is applied inconsistently | `sdlc-claude-md` | | The agent can't tell whether its own work is correct | `sdlc-feedback-loop` | | Changing the config that steers the agent, or the model | `sdlc-agent-evals` | | A rule must hold every time, or a named human must approve | `sdlc-hooks` | | More work in flight than one session can carry | `sdlc-parallel-work` | | PRs queueing, or review depth varying by reviewer | `sdlc-pr-review` | | Production signal should start work with nobody in the path | `sdlc-close-loop` | ## Why the chain Generating code got cheap. Everything bracketing it — agreeing what to build, judging whether the result is right, getting it shipped — did not. So those become the constraint, and the old controls stop fitting: reading every line was reasonable when a person typed every line. Replacing the controls with nothing is not the answer. Replacing the *enforcement* is. Each step leaves behind a file the next step reads: ``` intent the problem, in the originator's words sdlc-intent ↓ spec what to build, contradictions surfaced sdlc-spec ↓ plan files, seq