CLI Generator
The published @structor-dev/cli package inspects a local workspace, previews a setup transaction, and writes a sibling harness only after explicit approval.
Flagship Engineering Deep Dive
Structor
Repository-local AI harnesses for deterministic agentic development.
Structor is a local Harness Engineering Framework that generates repository-local AI engineering harnesses for software teams using Codex, Claude Code, and similar coding agents. Instead of treating AI prompts as scattered notes, Structor creates a versioned policy layer: context routing, contracts, validation scripts, review expectations, model overlays, and agent entrypoints that live beside the code they govern.
PublicDeveloper ToolingOpen source
Developer ToolingAI GovernanceAgentic DevelopmentLocal-First InfrastructureCLI Framework
Most tools organize prompts. Structor makes AI-assisted development more legible by giving repositories a local, reviewable operating surface for humans and agents.
Problem
AI coding agents are powerful, but most teams still manage them with loose prompts, tribal knowledge, and fragile convention. That creates drift: agents miss architecture boundaries, ignore review expectations, overwrite local guidance, or produce changes that are difficult to validate. Structor treats AI-assisted engineering as an infrastructure problem: make repositories legible to agents and humans through a durable, reviewable harness.
System Shape / Architecture
Generated Harness
The harness owns ai/ guidance, contracts, model overlays, review expectations, validation scripts, and generated reference structure while application repos keep product code.
Thin Entrypoints
Root AGENTS.md and CLAUDE.md files stay small and route agents back into canonical harness policy instead of becoming scattered policy stores.
Local Validation
Checks cover config, schemas, templates, tasks, contracts, model overlays, placeholders, examples, docs, package hygiene, smoke tests, and contributor setup.
System Flow
workspace/ product-frontend/ and product-api/ own application code product-structor/ owns AI engineering policy ai/PRODUCT.md, ARCHITECTURE.md, QUALITY.md, WORKFLOW.md ai/contracts/, ai/tasks/, ai/model-overlays/, ai/views/ scripts/check-workspace.mjs, check-overlay-drift.mjs, validate-governance.mjs consumer AGENTS.md and CLAUDE.md route into the generated harnessTechnical Highlights
- Built as a published npm CLI package: @structor-dev/cli.
- Generates repository-local AI harnesses for Codex and Claude Code.
- Creates thin AGENTS.md and CLAUDE.md entrypoints that route into canonical ai/ guidance.
- Supports governance contracts, model overlays, validation scripts, task templates, review guidance, and workspace checks.
- Preserves existing local guidance as source material instead of silently deleting or merging it.
- Separates deterministic setup completion from guidance readiness, so repo-specific policy migration remains a reviewed task.
- Explicitly avoids acting as an agent runner, hosted service, telemetry collector, PR bot, package installer, or LLM API wrapper.
AI / Automation Design
Structor assumes that AI coding agents need structured operating context, not just better prompts. The generated harness gives agents durable context about product intent, architecture boundaries, quality expectations, workflow rules, review behavior, and model-specific overlays. The AI layer is intentionally file-based and repository-local, which makes it inspectable in code review, compatible with different agent clients, usable without a hosted control plane, and deterministic enough to validate locally or in CI.
Engineering Tradeoffs
Structor deliberately does less than an orchestration platform. It does not run agents, call LLM APIs, open pull requests, install dependencies in consumer repos, collect telemetry, or mutate external services. That boundary keeps the framework safer and easier to adopt: it improves the operating environment around AI-assisted development without becoming another runtime dependency.
Stack
Runtime / CLI
Node.jsJavaScript ES modulesnpm package distribution
AI Engineering Surface
Codex entrypointsClaude Code entrypointsRepository-local governance filesModel overlaysContract manifests
Validation
Node test runnerDeterministic validation scriptsSchema checksTemplate checksContract checksSmoke tests
Documentation / Product System
Docs corpusGenerated file referencesRoadmapSecurity policyContribution flow
What This Demonstrates
Structor demonstrates my ability to design developer infrastructure around ambiguous, fast-moving AI workflows. It combines product thinking, CLI design, repository architecture, validation strategy, documentation systems, and practical AI governance. It is the project that best represents how I think about AI-native engineering: make the system legible, constrain the failure modes, and give both humans and agents a shared operating surface.
Next Technical Steps
- Add richer architecture diagrams for generated harness layouts.
- Expand examples for multi-repo product workspaces.
- Add migration tooling for existing agent guidance.
- Improve contract maturity levels.
- Strengthen validation reports for CI use.
Source Grounding
- README.md defines Structor as a local Harness Engineering Framework and states that it is a generator, not a runner.
- docs/concepts/what-structor-is.md and docs/concepts/harness-vs-runner.md document the harness boundary and non-runner posture.
- docs/reference/generated-files.md describes generated ai/ guidance, contracts, task templates, model overlays, review guidance, and validation scripts.
- package.json exposes the @structor-dev/cli package, structor bin, Node runtime, and validation/test script surface.
- docs/public-launch.md documents the public positioning and cautions against implying hosted services, telemetry, dashboards, or managed execution.