// Detail or invent
Missing context — or plausible fabrication.
Spec-Driven DevelopmentAIDDbot
Agents fail at scale
AI-Driven Development (AIDD) gives you structure, standards, and verification.// Guide or chaos
Code that ignores your standards.
Rules over Tools// Verify or hope
Silent drift until fixes are expensive.
Human in the LoopWhat you get
AIDDbot is a set of production-ready agent skills useful at the three stages of SDLC// Architect
Set up new projects and understand brownfield legacy codebases.
Architect skills// Builder
A guided path from requirement to verified solution delivery.
Builder skills// Craftsman
Review, repair, and release maintainable code with confidence.
Craftsman skillsWho is AIDDbot for
Developers and teams that want AI acceleration without sacrificing quality.// Frustrated by plausible-but-wrong code
When agents invent context or miss critical details, specs and verification keep output trustworthy.
// Standards that actually matter
Projects where consistency, conventions, and verifiability are non-negotiable.
Works in your environment
Use the same skills in the editors and agents you already rely on.// Antigravity · Cursor · VS Code
AI-powered IDEs that load project context from your repo.
// Claude Code · Codex · Copilot · OpenCode
Agent harnesses that invoke markdown skills from `.agents/skills/`.
Spec-driven development
Replace guesswork with an explicit spec the agent must satisfy before code exists.// Intent before implementation
When requirements live only in chat history, models interpolate and invent. AIDD treats the spec as the contract: scope, constraints, and acceptance criteria stay in-repo so every change traces back to written intent.
// Brownfield and greenfield
Specs are not only for new features. Capture current behavior, risks, and migration steps so refactors and legacy work stay grounded instead of drifting into silent rewrites.
Rules over tools
Project standards beat clever one-off prompts — especially when the whole team ships through agents.// Living instructions in the repo
AGENTS.md, skills, and conventions give the same guardrails in Cursor, Claude Code, Copilot, or the next editor. Tools change; the rules your team cares about stay versioned with the code.
// Consistency at scale
Without shared rules, every developer gets a different “voice” from the model. Central patterns for naming, architecture, and review keep output predictable enough to trust in production.
Human in the loop
Automation accelerates work; humans still own correctness, security, and product judgment.// Verify, then merge
Agents can propose diffs quickly — the bottleneck is knowing they are right. Build habits around tests, static checks, and targeted review so acceleration does not become silent debt.
// Catch drift early
Small mistakes compound when no one reconciles output against the spec. Short feedback loops (local runs, CI, and explicit sign-off) keep fixes cheap and intent aligned with what shipped.