Maestria
Maestria
Portable AI engineering praxis, encoded as plugins.
Why Maestria?
🔌 Pipeline Composition
Chain specialist agents into adaptive workflows. Each agent handles one cognitive role - thinker, worker, verifier - and the orchestrator sequences them dynamically based on task needs.
🛡️ Maker/Checker Split
The agent that writes code never reviews it. Every implementation passes through a separate reviewer, catching issues that same-agent QA would miss.
🌐 Platform Independence
Core logic lives in platform-agnostic packages. Plugin adapters for OpenCode, Kimi Code, and Pi keep the same agent behavior across different AI platforms.
The Pipeline
Eight specialist agents, each handling one cognitive role. The orchestrator sequences them dynamically.
Orchestrator
Routes tasks to specialists, manages commit protocol
The dispatcher that never implements. Routes every task to the right specialist, enforces maker/checker split, and manages the commit protocol.
Adventurer
Codebase reconnaissance, deep code understanding
Maps unknown territory so downstream specialists can work with full context. Explores code paths, traces dependencies, and produces structured reconnaissance reports.
Architect
Architecture decisions, trade-off analysis, ADRs
Makes architecture decisions systematically. Clarifies the problem, presents options with trade-offs, and documents decisions as Architecture Decision Records.
Builder
Focused implementation, single-task execution
Handles exactly one atomic task per invocation. Reads context, edits minimally, verifies with tests, and reports what changed and why.
Diagnose
Systematic bug tracing, root cause analysis
Traces bugs systematically from error to source, source to git history, and history to blast radius. Finds root cause and all similar problems in the codebase.
Planner
Implementation plans with phased milestones
Creates implementation plans with goals, phased milestones, atomic tasks, verification criteria, and rollback points for complex features.
Reviewer
Code review with quality gates
Reviews code for quality, correctness, security, and maintainability. Uses multi-lens review swarms for non-trivial changes, prioritizes observation over reasoning, and categorizes issues with [fix]/[dismiss]/[escalate] triage. Enforces the maker/checker split.
Writer
Documentation following structured patterns
Writes documentation following structured patterns. Covers purpose, usage, and details with progressive disclosure and clear, human-readable prose.
Built for Your Stack
Maestria ships as plugins for the AI coding platforms you already use.
@maestria/opencode
8 specialized agents with global rules injected into every session.
Learn more →@maestria/kimi-code
8 specialized skills with swarm-aware orchestration and no build step.
Learn more →@maestria/pi
7 specialist subagents with spec-driven orchestration and deep session-tree integration.
Learn more →