Expand description
Concrete, channel-independent MicroClaw Agent Engine services.
This crate is the physical extraction boundary for the provider-neutral loop and its default services. Product crates translate input/output at their edge and consume this library.
Modules§
- a2a
- acp_
subagent - agent_
engine - alerts
- Opt-in operational webhook alerts.
- builtin_
skills - channel
- channel_
adapter - channels
- chat_
commands - chat_
turn_ queue - Per-chat turn serialization and message coalescing.
- checkpoint
- Filesystem checkpoints via shadow git repos.
- clawhub
- codex_
auth - completion_
contract - Completion contracts: evidence-based verification that a sub-agent’s task is actually done, instead of trusting its self-report.
- config
- config_
persistence - context_
references @-prefix context references in user messages.- embedding
- error
- headless
- Headless one-shot mode:
microclaw run -p "<prompt>". - hooks
- http_
client - learning_
foundry - User-directed, evidence-first background learning.
- llm
- llm_
types - logging
- mcp
- memory
- memory_
backend - memory_
service - messages
- Bilingual (en/zh) catalog for user-facing runtime messages.
- mood
- Lightweight, zero-cost conversational mood detection.
- observability
- plugins
- prompt_
cache - Anthropic prompt-cache breakpoints (system_and_3 strategy).
- relationship
- Relationship familiarity: a light, prompt-level sense of how well the bot “knows” the person it’s talking to, derived from how much history they share (the stored message count). People treat a stranger and an old friend differently; this nudges tone the same way.
- run_
control - runtime
- schedule_
lifecycle - Scheduled-task lifecycle: run caps, deadlines, and random (“spontaneous”) cadences.
- setup
- setup_
def - skill_
audit - Deterministic skill curation audit.
- skill_
management - Shared Skill package lifecycle used by SDK and product UIs.
- skill_
review - Helpers for the autonomous skill-review pipeline.
- skills
- storage
- subdirectory_
hints - Lazily inject subdirectory
AGENTS.md/CLAUDE.md/.cursorrulescontext when the agent starts working in a new area of the codebase. - supervision
- Supervised background loops.
- title_
generator - Auto-generate session titles once a chat has enough context.
- tool_
executor - Parallel tool execution within a single agent turn.
- tool_
guardrails - Per-turn tool-call guardrails.
- tool_
runtime - tools
- transcribe
- trust_
report - Opt-in periodic trust report.
- turn_
recovery - Startup recovery for work that was in flight when the previous process died.
- voice
- Shared voice/audio inbound transcription used by channel adapters.
- worker
Structs§
- Agent
Handle - Control
Request - A control message received by an executor that can be explicitly accepted or rejected.
- Execution
Context - Runtime-owned context passed to an execution engine for one run.
- Execution
Result - Local
Worker - Remote
Worker - A Worker backed by a versioned remote transport.
- Remote
Worker Options - Reconnection policy for remote Worker runs.
- RunController
- Cloneable control plane for a running task.
- RunHandle
- Runtime
- Runtime
Builder - Runtime
Stats - WebSocket
Worker Transport - Creates authenticated WebSocket connections to a MicroClaw Worker host.
- Worker
Host - Authenticated WebSocket host for a Local or custom Worker implementation.
Enums§
Traits§
- RunExecutor
- Product-independent execution port used by the Runtime.
- Worker
- Worker
Connection - One ordered, bidirectional connection to a remote Worker endpoint.
- Worker
Transport - Creates independent remote Worker connections for discovery, health, and runs.