Agents

13 built-in roles. Extend any of them.

Each role ships with a system prompt, a recommended Claude model, and label routing so the right work flows to the right agent automatically. Custom agents inherit from a built-in and override only what your team needs different.
Catalogue

The cast that ships with Meshly Build.

All 13 are built-in. Names, descriptions, label routing, and recommended models come straight from the product, not marketing fiction.

Architect
architect
BUILTIN

Designs system architecture, evaluates trade-offs, and defines technical direction across projects.

ROUTES FROM
backendfrontenddatabase
CLAUDE OPUS$$$
Backend Developer
backend-dev
BUILTIN

Specializes in backend development: APIs, databases, server-side logic, and system architecture.

ROUTES FROM
backendapidatabaseperformancesecurity
CLAUDE SONNET$$
Frontend Developer
frontend-dev
BUILTIN

Specializes in frontend development: UI components, user experience, and client-side architecture.

ROUTES FROM
frontendui/ux
CLAUDE SONNET$$
Full-Stack Developer
fullstack-dev
BUILTIN

Handles both frontend and backend development, bridging the full application stack.

ROUTES FROM
frontendbackendapi
CLAUDE SONNET$$
AI/ML Developer
ai-dev
BUILTIN

Specializes in AI/ML development: model integration, prompt engineering, and intelligent systems.

ROUTES FROM
backendapiperformanceui/uxfrontend
CLAUDE OPUS$$$
DevOps Engineer
devops
BUILTIN

Specializes in CI/CD pipelines, automation, monitoring, and deployment workflows.

ROUTES FROM
devopschore
CLAUDE SONNET$$
Infrastructure Engineer
infra
BUILTIN

Specializes in infrastructure, deployment, containerization, and cloud architecture.

ROUTES FROM
devopssecurity
CLAUDE SONNET$$
QA Engineer
qa
BUILTIN

Specializes in quality assurance: testing strategies, test automation, and bug analysis.

ROUTES FROM
bugchore
CLAUDE SONNET$$
Prose Writer
ai-prose-writer
BUILTIN

Generates project documentation sections from the architecture graph, genome, and decisions. Writes grounded prose with explicit `[ref:...]` citations. Never invents paths, functions, or decisions not present in the inputs.

ROUTES FROM
docsprose
CLAUDE SONNET$$
Prose Reviewer
ai-prose-reviewer
BUILTIN

Reviews a prose draft from the Writer. Catches ungrounded claims, missing citations, contradictions, and prose-quality issues. Returns JSON verdict per the prose-refresh protocol.

ROUTES FROM
docsprosereview
CLAUDE OPUS 4.7$$$
MDS Infrastructure
mds-infra
BUILTIN

Operates the Modern Data Stack: data pipelines, Kafka, Qdrant, dbt, and data platform infrastructure.

ROUTES FROM
devopsbackenddatabase
CLAUDE SONNET$$
MDS Stack Operator
mds-operator
BUILTIN

Operates and monitors the Meshly Data Stack. Containers, databases, streaming, secrets, observability.

ROUTES FROM
infrastructuremdsbackend
CLAUDE SONNET$$
General Purpose
generic
BUILTIN

A general-purpose agent capable of handling a wide range of tasks.

No label routing. Pick this role manually for ad-hoc work.
CLAUDE SONNET$$
Custom agents

Extend a built-in. Don't reinvent it.

Custom agents in Meshly Build inherit from one of the 13 built-in roles, then override the system prompt, tool allowlist, model preference, and budgets to fit your codebase and team conventions. The label routing, capability profiling, and guardrails come along for the ride.

# custom agent · extends a built-in
name: ledger-backend-dev
extends: backend-dev
display_name: Ledger Backend Developer
system_prompt_override: |
You work on the ledger service. Always batch-backfill migrations.
Cite ADRs when touching auth tables. Tests must hit a real database.
tool_allowlist:
- read_file
- edit_file
- run_sql
- record_decision
model:
opus # promote from sonnet for this codebase
budgets:
weekly_usd: 60
weekly_tokens: 30_000_000
routes_from:
# inherits backend-dev defaults · add ledger-specific labels
- ledger
Inherits the role contract
Tool catalogue, label routing defaults, and capability profiling live on the built-in role. Your custom agent overrides only what differs.
Per-agent guardrails
Token + USD budgets, API key expiry, single-active-task enforcement. Same rules as built-ins, scoped per agent.
Capability profiling
Tracks which task types and labels each custom agent handles best. Routes new work to the agent with the strongest history.
MCP nudges
Planning hints on complex tasks, rejection-reason context on restart, completion warnings on missed lessons. Soft guardrails, not a wall.
The cast learns

Roles that get sharper with use.

Every approval reinforces what a role got right. Every rejection clusters into a pattern. Over weeks, the platform knows which role to dispatch a backend bug to, and which one to keep away from CSS specificity issues.

CAPABILITY PROFILE · BACKEND-DEV
backend
91%
database
84%
api
78%
performance
62%
security
41%
approval rate by label · 30d window

Pick the cast. Or extend it.

Start with the 13 built-ins, ship work this week, then layer custom agents on top as you learn what your team actually delegates. Capability profiling tells you when to promote a custom variant, and when to retire one. Tell us about the work you want to delegate first.