Show HN: Outline Driven Development – New AI-Assisted Coding Paradigm; BN (github.com)

Outline Driven Development (ODD) is a new AI-assisted coding paradigm and open-source toolset that treats a versioned, content-addressable “outline” as the single source of truth for agentic code generation. The project provides CLI/plugins for Gemini, Claude and Codex and prescribes an “outline-as-assembly” workflow: human intent, compliance constraints and architectural guardrails are compiled into a hashed outline; every LLM invocation receives an outline slice plus explicit success metrics and must produce artifacts that revalidate against the outline or the step is halted/replayed. The stack mandates a battery of deterministic tooling (lsd, ast-grep, ripgrep, fd, LangGraph, MCPs) and embeds telemetry, test verdicts and rubric scores into a governance loop that refines outlines over time. For the AI/ML community this addresses reproducibility, safety and auditability of non-deterministic LLM outputs by enforcing contracts, budgets (latency/memory), pre/postconditions and strict schema/versioning. Key technical features: content-addressable outline storage with monotonic hashes, outline slices for minimized agent surface, deterministic replay harness comparing filesystem/output digests, diff-noise thresholds (≤2% target) that trigger outline tightening, fail-closed behavior for missing telemetry or tool drift, chaos-resilience tests, and append-only audit logs with rollback hooks. The approach makes LLMs first-class, bounded modules in an architecture-first pipeline suitable for regulated environments, strict QA gates, and reproducible engineering workflows.