Loading…
Loading…
Features · Scale · Deep agents
Deep agents add autonomy on top of the loop: a GoalAgent works toward success criteria, a ReflectiveAgent critiques and revises, a Supervisor manages workers, and a ShipCrew runs tasks with dependencies. All support MCP, built-in tools, streaming and memory.
from shipit_agent import Goalfrom shipit_agent.deep import GoalAgentgoal = Goal( objective="Add input validation to the signup endpoint", success_criteria=["rejects empty emails", "has tests", "tests pass"],)
Goal
objective and criteria
Plan
decompose the work
Act and check
criteria met?
Done
or revise again
The code is a checked example from the docs; the output is an example of what it prints.
from shipit_agent import Goal
from shipit_agent.deep import GoalAgent
goal = Goal(
objective="Add input validation to the signup endpoint",
success_criteria=["rejects empty emails", "has tests", "tests pass"],
)
agent = GoalAgent.with_builtins(llm=llm, goal=goal)
for event in agent.stream():
print(event.payload.get("output", ""))6 more ways to use it
Each example is from the docs and checked against the library.
That gives you the seven deep tools, the full built-in tool catalogue, and the opinionated `DEEP_AGENT_PROMPT`.
from shipit_agent.deep import DeepAgent
agent = DeepAgent.with_builtins(llm=my_llm)
result = agent.run("Investigate the auth bug and propose a fix.")
print(result.output)GoalAgent
Works toward success criteria and checks them before it stops.
ReflectiveAgent
Drafts, critiques and revises until it meets the bar.
Supervisor
A lead that hands work to specialists and merges it.
ShipCrew
Tasks with dependencies; independent ones run in parallel.