Skip to content

maestria

OPEN-SOURCE AGENT METHODOLOGY

Make every step intentional.

A portable harness for AI coding agents. Route complex work through specialists, structured handoffs, and independent review.

npx maestria install opencode

npm · pnpm · yarn · bun · deno
- any runtime works

Harness transcript example

maestria - harness transcriptEXAMPLE

npx maestria install opencode

@maestria/opencode installed - 8 agents wired

maestria status

opencodeconnectedclaude-codeconnected
kimi-codeavailablecursoravailable

# one task through the harness - thinker, worker, verifier

thinkeradventurerrecon report
thinkerarchitectdecision record
thinkerplannerphased plan
workerbuilderimplementation + tests
verifierreviewerverdict: approved

01 / THE METHOD

Structure beats improvisation.

01

Pipeline composition

Route work through focused stages so no single context does everything.

02

Maker/checker split

The builder produces; the reviewer sees it fresh.

03

Platform independence

The methodology stays consistent while each platform uses its native primitives.

02 / SPECIALISTS

One dispatcher, seven specialists.

Orchestrator

Routes tasks to specialists, manages commit protocol

The dispatcher selects the smallest safe route for each turn: it executes known low-risk work directly, delegates specialist work with concise briefs, integrates results, enforces the maker/checker split, and manages the commit protocol.

Scope
Routes work; executes direct-route tasks itself
Artifact
Route assignments, commit protocol
Constraint
Maker/checker split enforced
Full specialist reference

03 / ADAPTERS

One methodology, built on native primitives.