v5.28.0
2 次代码提交
| 作者 | SHA1 | 备注 | 提交日期 | |
|---|---|---|---|---|
 Asim Aslam
|
e416ea4a75 |
Enhance agent workflows with guardrails and documentation updates (#2952)
* docs: map go-micro onto Anthropic's workflows-vs-agents taxonomy - new guide 'Agents and Workflows': adopts Anthropic's Building Effective Agents vocabulary — workflow (predefined path) = flow, agent (dynamic self-direction) = agent — maps the augmented-LLM building block and the five workflow patterns onto go-micro, and shows routing (chat router) and orchestrator-workers (conductor + plan/delegate) are already native. - flow package doc reframed as a workflow (predefined path) per the same taxonomy, with guidance on flow vs agent. - nav + README link the new guide. * feat: agent guardrails — step limit and tool approval hook Anthropic's Building Effective Agents stresses stopping conditions and human-in-the-loop checkpoints for autonomous agents. Add both as plain options enforced at the tool-handler choke point — no provider changes, no new abstraction: - MaxSteps(n): bound tool executions per Ask; beyond the limit, actions are refused and the model is told to stop and summarize. - ApproveTool(fn): gate each action before it runs; returning false blocks it and surfaces the reason to the model. The internal plan tool is never gated. Exposed at the micro package (AgentMaxSteps, AgentApproveTool, ApproveFunc). Tests cover the limit, blocking, and that plan is not gated. Guardrails section of the agents-and-workflows guide updated from 'active work' to documented options. * feat: flow can dispatch to an agent (flow triggers, agent reasons) Unify the engine without collapsing the workflow/agent distinction. A Flow with Agent set hands each event's rendered prompt to a named registered agent over RPC (Agent.Chat) instead of running its own LLM step — so the workflow stays the deterministic trigger and the agent is the reasoning engine, with its plan, delegate, memory, and guardrails. A plain flow is unchanged (single augmented-LLM step). - flow.Agent(name) / micro.FlowAgent(name); flow stores the client and skips model setup when dispatching. - test: dispatch routes to comms.Agent.Chat with the rendered prompt and records the reply. - guide: 'Flow triggers, Agent reasons' section. --------- Co-authored-by: Claude <noreply@anthropic.com> |
||
|
Asim Aslam
|
cb33decd97 |
Add built-in plan and delegate tools for agents with examples (#2949)
* feat: add plan and delegate as built-in agent tools Give agents two self-capabilities, expressed as plain tools wired into the existing tool handler — no harness or graph, consistent with "services are the only abstraction": - plan: record/update an ordered plan, persisted to store-backed memory and surfaced in the system prompt on later turns (externalized planning). - delegate: hand a self-contained subtask to another agent. Delegate-first — if the target names a registered agent it is called via RPC; otherwise a focused ephemeral sub-agent is created with agent.New + Ask in a fresh, isolated context (loads/persists no history, no built-in tools, so it cannot re-delegate). Both are added automatically to any non-ephemeral agent, so existing micro.NewAgent services and micro chat routing get them for free. Tests are hermetic (memory store + memory registry). * feat: add agent-plan-delegate example and document plan/delegate - examples/agent-plan-delegate: coordinator that plans multi-step work, creates tasks with its own tools, and delegates notification to a separate registered comms agent over RPC. - integration tests driving the full Ask loop through a fake provider: plan tool exposure + persistence, ephemeral delegation with isolated context, delegate-first RPC routing to a registered agent. - docs: README (Building Agents + features + examples), AGENT_DESIGN (Built-in Capabilities), agent-patterns guide (Pattern 9), CLAUDE.md. * docs: blog post and guide for plan & delegate - blog/17: "Plan & Delegate: Deep Agents in Go" — what the feature is, how plan and delegate work, and a runnable getting-started path. - guides/plan-delegate: reference guide with the smallest-agent snippet, plan/delegate semantics, and the multi-agent example; linked in nav. - example: auto-detect provider/key from common env vars (ANTHROPIC_API_KEY, OPENAI_API_KEY, ...) so 'export KEY && go run main.go' just works. - onboarding: getting-started paths now include go mod init / go get and a clone-and-run path, so a reader can actually run it from a cold start. * refactor: reframe plan/delegate blog and clean up sub-agent construction - blog/17 retitled "Agents That Plan and Delegate" and reframed around intent (plan = state intent, delegate = direct it), positioned as the next beat after blog 15/16 and tied to the existing store + agent RPC rather than re-announcing them. "Deep agents" now a single in-passing nod, matching how blog 14 references LangChain. - agent: add unexported newEphemeral constructor for sub-agents instead of type-asserting the public Agent interface to set an internal field; matches the options-only construction idiom used elsewhere. * feat: expose plan & delegate in the micro chat fallback Add agent.Builtins(opts...) — returns the built-in tools plus a handler, so the plan/delegate capabilities can be wired into a tool loop that isn't a running Agent. micro chat's direct-service fallback now reuses it (single source of truth, no duplicated handler logic), so planning and delegation are available there too, not just for registered agents. Adds a test for the accessor; notes CLI availability in the guide. --------- Co-authored-by: Claude <noreply@anthropic.com> |