A local-first engineering workflow for AI-assisted development that keeps intent, context, verification evidence, and human review connected from task to PR.
Maverick makes the work around a change visible: the problem, relevant context, plan, evidence, and review decision.
Small briefs say what changes, what does not, and how completion is verified.
Targeted paths and local conventions replace repository-wide guessing.
Evidence and the diff support the decision to merge.
$ npx maverick-ai-dev-stack init ✓ project initialized: .maverick/config.json $ npx maverick-ai-dev-stack task add-pagination ✓ task created: .maverick/tasks/add-pagination $ npx maverick-ai-dev-stack validate add-pagination PASS required artefacts and sections are present (warnings may remain).
Problem, goal, constraints, and verification make intent reviewable.
Architecture, paths, rules, and commands give an agent a bounded start.
Tests, security checks, and a human decision close the loop.
Claude Code, Codex, Cursor, and generic agents can read the same artefacts. The Lab helps teams test a workflow carefully; it does not manufacture benchmark claims.
An agent can produce a diff quickly; engineering confidence comes from making the intent, relevant context, scope, checks, and review decision inspectable. Maverick is the workflow around the coding agent, not a competing agent or model.
Without: prompt → agent → diff → “looks good”
With: intent → context → plan → build → verify → review → PRInspect tasks, documentation, Git state, available commands, and workflow signals from a local dashboard—without uploading the repository. Start with maverick dashboard --open, then read the Lens guide or the feature announcement.
npx maverick-ai-dev-stack init
npx maverick-ai-dev-stack map
npx maverick-ai-dev-stack task add-orders-pagination --preset standard
npx maverick-ai-dev-stack packet add-orders-pagination --agent codexContinue with Getting started. AI writes; engineers decide.