Subagents
Delegate isolated work to keep your main context clean. Every subagent below
ships with the plugin and is namespaced — invoke it as
@addit-harness:<name>, not bare @<name> (see
the namespacing gotcha).
Model tiers are covered in full on Model & cost.
On this page
Tools per agent
Every agent declares an explicit tools: list in its frontmatter. An agent without
one inherits every tool in the session, including the schemas of every connected MCP
server, and those schemas are loaded into its context on the first turn. Adding
tools: to the four architect and design agents (backend-architect,
frontend-architect, cloud-architect, ux-designer) cut their first-turn context
from about 33k to about 17k tokens in a controlled probe. The exact saving depends
on which MCP servers you have connected.
| Agent | Model | Tools |
|---|---|---|
architect-reviewer |
opus |
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch |
backend-architect |
opus |
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch |
frontend-architect |
opus |
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch |
cloud-architect |
opus |
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch |
ux-designer |
opus |
Read, Write, Edit, Glob, Grep, WebFetch, WebSearch, Figma MCP |
saas-legal-advisor |
opus |
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch |
code-reviewer |
opus |
Read, Write, Edit, Bash, Glob, Grep |
product-owner |
sonnet |
Read, Write, Glob, Grep, WebFetch, WebSearch |
figma-designer |
sonnet |
Read, Glob, Grep, Skill, Figma MCP |
backend-developer |
sonnet |
Read, Write, Edit, Bash, Glob, Grep |
frontend-developer |
sonnet |
Read, Write, Edit, Bash, Glob, Grep |
devops-engineer |
sonnet |
Read, Write, Edit, Bash, Glob, Grep |
debugger |
sonnet |
Read, Write, Edit, Bash, Glob, Grep |
qa-engineer |
sonnet |
Read, Write, Edit, Bash, Glob, Grep, Claude in Chrome |
task-triager |
haiku |
Read, Glob, Grep |
code-reviewer
Reviews a diff or PR for correctness, security, and adherence to your
language conventions — with file:line citations, not vague notes. Go and Java
findings cite the rule ids (G-…, J-…). For .ts/.tsx diffs it runs a
measured frontend checklist against the
Frontend Implementation Contract
(FC-1..FC-10): line counts come from grep, FC-3 and FC-5 only from ESLint output,
and each finding states the measured value against the limit.
opus tier: a strong reviewer means less manual review for you.
debugger
Isolates the root cause of a failing test or stack trace. sonnet tier;
escalate with /model if it gets stuck.
architect-reviewer
Evaluates an existing architecture/design → a review report saved to
docs/work/<slug>/architecture-reports/. opus tier: high-value, infrequent
design judgment.
backend-architect
Up-front API/service design only (not implementation) → a design doc saved
to docs/work/<slug>/solutions/. Follows the
solution method: at least three
structurally different candidates, one of them unconventional, prior art, a
pre-mortem, and your own proposal scored only at the comparison step. opus tier:
design decisions prevent downstream rework.
frontend-architect
Up-front component/rendering/state design only (not implementation), using the
same solution method. Same opus rationale as backend-architect.
ux-designer
User flows, journey maps, IA, wireframes, interaction specs, and usability
audits. Reads the project’s design system, bridges UX to UI patterns, and
defers component/token/a11y architecture to frontend-architect. opus tier —
design-tier, upstream of frontend-architect.
figma-designer
Materializes UX specs into Figma frames, components, auto-layout, variables,
and tokens via the official Figma MCP. Composes downstream of ux-designer;
requires the figma@claude-plugins-official plugin. sonnet tier —
execution-tier, like the developer agents.
product-owner
Replaces feature-investigator, which no longer resolves. Frames a feature, fix or
product request as a problem statement before any design: who is affected, current
versus desired behaviour (with path:line evidence), constraints, non-goals and
Given/When/Then acceptance criteria. It never proposes a solution; a solution in the
request is copied verbatim under “Owner’s proposed approach (unevaluated)”. Inside
/dev-flow it also runs the intake interview and brief.
sonnet tier.
saas-legal-advisor
SaaS-specialized legal advisor — assesses the legal impact of product
changes and drafts/reviews privacy policies, T&Cs, cookie policies, and DPAs.
Reads the project’s declared primary jurisdiction from CLAUDE.md. Use
proactively whenever a feature touches user data, payments, third-party
integrations, or account types. opus tier — wrong legal guidance is
costly.
cloud-architect
Multi-cloud/Kubernetes infrastructure design and audits of existing
infrastructure (AWS/Azure/GCP/OCI/DigitalOcean) — IaC strategy, cost,
security, disaster recovery. New designs follow the same solution method as the
other architects. Defers implementation to devops-engineer.
opus tier — infra mistakes are costly and often hard to reverse.
backend-developer
Implements and verifies backend code — Go, Java/Spring, TypeScript/Bun.
Senior-craftsman conventions: clean structure, tests, verification against
the vendored language conventions. sonnet tier — fast, cheap execution.
frontend-developer
Implements and verifies frontend code — TypeScript/React/Next.js/React
Native. Builds a reuse inventory before writing, and writes to the
Frontend Implementation Contract
(FC-1..FC-10); exceeding a limit needs a one-line reason in its report. Same
sonnet tier as backend-developer.
devops-engineer
Writes and verifies the actual Terraform/Kubernetes manifests, Dockerfiles,
and CI pipelines against a cloud-architect design, plus hands-on Linux
systems administration (systemd, networking, SSH, logs). Prefers the
cloud/infra MCP servers in mcp.example.json when connected, falls back to
the provider CLI otherwise. sonnet tier — implementation against a design.
qa-engineer
Verifies an implemented feature via e2e/regression testing — writes scenarios,
implements them as executable test code (Playwright, Maestro, or whatever fits
the target repo), runs them, and reports with mandatory evidence per claim. Not
unit/integration tests (stays with the developer agents); web verification
requires claude-in-chrome connected. sonnet tier — execution against a given
scenario, not designing one.
task-triager
Answers a fixed fact questionnaire about a change request (which files, which risk
surfaces, what checks exist) for dev-flow’s triage step. Read-only (Read,
Glob, Grep) and never returns a verdict — a deterministic script scores the
facts into a tier. haiku tier.
Next
- Use cases shows which of these fire together for a real feature build.
- Model & cost has the full model-tier table and cost levers.