Documentation
¶
Overview ¶
Package subagent runs a specialized agent loop in isolation so a primary agent can delegate a self-contained task and receive only the final result back.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ErrMaxDepth = errors.New("subagent: max delegation depth exceeded")
ErrMaxDepth is returned when subagent nesting exceeds the configured delegation depth.
var ErrNoModel = errors.New("subagent: model is required")
ErrNoModel is returned when a subagent run is requested without an LLM model.
Functions ¶
func FinalText ¶
func FinalText(messages []agent.AgentMessage) string
FinalText returns the concatenated text of the last assistant message in the run.
Types ¶
type Definition ¶
type Definition struct {
// Name is the unique subagent identifier referenced by the delegate_subagent tool.
Name string
// Description tells the primary agent when to delegate to this subagent.
Description string
// SystemPrompt is the isolated instruction set for the subagent loop.
SystemPrompt string
// Tools restricts which registered tools the subagent may call.
Tools []string
// Skills restricts which agent skills are injected into the subagent prompt and read_skill allowlist.
Skills []string
// Model optionally overrides the model used for the subagent loop.
Model string
}
Definition describes a specialized subagent that the primary agent can delegate to. Tools is an allowlist of tool names exposed to the subagent; empty means all tools registered on the provided registry.
type Deps ¶
type Deps struct {
// Model is the resolved LLM client for the subagent loop.
Model llms.Model
// Registry holds the tools available to the subagent after allowlist filtering.
Registry *tool.Registry
// Config is the loop configuration; defaults are applied when zero.
Config agent.Config
// Depth is the current delegation depth of the caller (0 for the primary agent).
Depth int
// MaxDepth caps nested delegation; values <= 0 default to 1.
MaxDepth int
}
Deps carries the runtime dependencies required to run a subagent loop in isolation.
type ProgressFn ¶
type ProgressFn func(update string)
ProgressFn receives human-readable progress updates from the subagent run.
type Result ¶
type Result struct {
// Text is the concatenated assistant text produced by the subagent.
Text string
// Messages are all messages generated during the subagent run.
Messages []agent.AgentMessage
}
Result is the outcome of a completed subagent run.
func Run ¶
func Run(ctx context.Context, def Definition, deps Deps, prompt string, onProgress ProgressFn) (Result, error)
Run executes a subagent in a fresh, isolated context and returns its final assistant text. The run does not share the parent conversation and is not persisted to any session tree, matching the mainstream subagent contract.
Source Files