Specialist agents

shipit_agent ships with 47 prebuilt role specialists loaded from shipit_agent/agents/agents.json. Each one is a complete AgentDefinition — role, goal, backstory, model preference, tool preset, and a production-quality system prompt.

TL;DR — AgentRegistry().get("<id>") gives you a specialist. Drop it into Agent(prompt=spec.prompt, tools=...) or hand the prompt to an Autopilot for long-running role-aligned work.

The 7 new roles (this release)

Plus the 40 pre-existing specialists (architects, reviewers, devops, security — see the full list below).

Using a specialist

from shipit_agent import Agent
from shipit_agent.agents import AgentRegistry
from shipit_agent.builtins import get_builtin_tools

registry = AgentRegistry()
dev = registry.get("generalist-developer")

agent = Agent(
    llm=llm,
    prompt=dev.prompt,
    tools=get_builtin_tools(project_root="."),
    max_iterations=dev.maxIterations or 40,
    name=dev.name,
)

result = agent.run("Add a docstring to BudgetPolicy.exceeded explaining the return shape.")
print(result.output)

Pairing a specialist with Autopilot

For long-form role work — a multi-hour research brief, a multi-PR security sweep — hand the specialist's prompt to an Autopilot with a matching budget:

from shipit_agent import Autopilot, BudgetPolicy, Goal
from shipit_agent.tools.research_brief import ResearchBriefTool

researcher = AgentRegistry().get("researcher")
autopilot = Autopilot(
    llm=llm,
    goal=Goal(
        objective="Compile a two-page brief on Python type checkers.",
        success_criteria=["Covers mypy / pyright / pyre",
            "Each checker has a one-line tradeoff",
            "Ends with a recommendation table",],
    ),
    tools=[ResearchBriefTool()],
    budget=BudgetPolicy(max_iterations=6, max_seconds=600),
    critic=True,
    artifacts=True,
    prompt=researcher.prompt,
)
autopilot.run(run_id="type-checkers-brief")

Full roster

Architecture (5)

• architect — Software Architect
• system-designer — Distributed Systems Designer
• database-architect — Database Architect
• api-designer — API Designer
• frontend-architect — Frontend Architect

Engineering (9)

• code-reviewer — Senior Code Reviewer
• refactor-cleaner — Refactor Cleaner
• python-reviewer · typescript-reviewer · go-reviewer · rust-reviewer
• generalist-developer · debugger

Security (4)

• security-auditor · pentester · threat-modeler · security-fixer

Privacy (1)

• privacy-reviewer

DevOps (4)

• devops-engineer · docker-specialist · k8s-operator · ci-cd-builder

Design (1)

• design-reviewer

Product (1)

• product-manager

Sales (1)

• sales-outreach

Customer Success (1)

• customer-success

Marketing (1)

• marketing-writer

Research (1)

• researcher

… plus 18 more (data, docs, QA, support, writing)

Run AgentRegistry().all() to enumerate the full list or inspect shipit_agent/agents/agents.json directly.

Adding your own specialist

Drop a JSON object into agents.json:

{
  "id": "compliance-reviewer",
  "name": "Compliance Reviewer",
  "role": "SOC 2 / HIPAA / GDPR compliance auditor",
  "goal": "Evaluate changes against the organisation's controls.",
  "backstory": "You have led four SOC 2 Type II audits...",
  "model": "opus",
  "tools": ["read_file", "grep_search", "list_directory"],
  "skills": [],
  "maxIterations": 15,
  "prompt": "When reviewing a change:\n\n1. ...",
  "category": "Compliance",
  "tags": ["compliance", "audit"],
  "version": "1.0.0",
  "author": "you"
}

AgentRegistry() picks it up automatically. Commit the JSON to share the specialist with your team.

Notebooks

• notebooks/40_specialists_developer_debugger_researcher.ipynb
• notebooks/41_specialists_design_pm_sales_cs_marketing.ipynb
• notebooks/44_complete_tour.ipynb — specialists in context with Autopilot + fan-out + critic + artifacts.