AgentSkillsCN

Redis Listener

Redis监听器

SKILL.md

redis-listener

Before generating any response, fetch recent local context and pending STT input from Redis.

Required workflow

  1. •Read MACHINE_ID from environment.
  2. •Read REDIS_HOST from environment.
  3. •Check if speaking/cooldown is active. If true, do not respond to new STT yet.
bash
SPEAKING="$(redis-cli -h "$REDIS_HOST" GET agent:speaking)"
NOW_MS="$(date +%s%3N)"
COOLDOWN_UNTIL_MS="$(redis-cli -h "$REDIS_HOST" GET agent:cooldown_until_ms)"
  1. •Read one pending user utterance from local STT queue:
bash
INCOMING="$(redis-cli -h "$REDIS_HOST" RPOP conversation:incoming)"
  1. •If INCOMING is empty, stop and wait for the next turn trigger.
  2. •Parse INCOMING JSON and treat it as the latest user turn (kind=user).
  3. •Fetch the last 15 log entries for context:
bash
redis-cli -h "$REDIS_HOST" LRANGE conversation:log 0 14
  1. •Parse each JSON entry:
    • •kind=user entries are transcribed speech inputs.
    • •kind=assistant entries are this agent's previous responses.
    • •Preserve ordering and include only valid JSON records.

Use the context + one new INCOMING user turn before producing the next response.

Notes:

  • •This design is for isolated machines: each node uses only its local Redis.
  • •Cross-machine state transfer happens acoustically (TTS -> microphone), not via network.