What problem does it solve? Building Lark/Feishu bots and real-time integrations normally requires standing up webhook servers and handling push callbacks. This Skill lets an AI agent subscribe to and consume Lark events (IM messages, reactions, chat changes, VC meeting end, Minutes generated, Whiteboard updates) directly as NDJSON streams from a CLI subprocess, with bounded runs and a machine-readable ready/exit contract. ## Core Features & Use Cases - Event discovery and schema inspection: List all subscribable EventKeys with lark-cli event list and inspect params, output fields, and jq root paths via event schema <key> --json before writing filters. - Streaming consumption with jq projection: Run lark-cli event consume <EventKey> to emit one JSON event per line on stdout, optionally filtered or transformed with --jq, and bounded with --max-events or --timeout. - Subprocess orchestration contract: A fixed stderr ready marker, stdin-EOF graceful shutdown, structured JSON error envelopes, and documented exit codes (0-5) make it safe for AI agents to spawn and supervise consumers. - Use Case: Run one consumer for im.message.receive_v1 and another for im.message.reaction.created_v1 concurrently, each writing NDJSON to its own file, to power a bot that reacts to group chat activity in real time. ## Quick Start Ask the agent to list available Lark event keys, then start consuming im.message.receive_v1 as the bot with a one-event limit to inspect the payload shape.