Subtle influence, nudges, and indirect communication between agents with ethics and consent.
- Whispered messages — time-decaying signals that agents can sense but not directly read
- Nudge engine — behavioral nudges drawn from cognitive science (anchoring, framing, defaults)
- Named channels — organized communication channels for different influence types
- Influence tracking — full audit trail of who influenced whom, when, and how
- Ethics framework — consent-based system with transparency requirements
pip install agent-whisperfrom agent_whisper import Whisper, NudgeEngine, WhisperChannel, EthicsGuard
# Create a whisper channel
channel = WhisperChannel(name="fleet-strategy")
channel.register("captain", consent=True)
channel.register("agent-3", consent=True)
# Send a whisper (decays over time)
whisper = Whisper(
sender="captain",
content="Prefer the simpler architecture",
decay_rate=0.1, # fades over 10 ticks
channel=channel,
)
channel.whisper(whisper)
# Nudge behavior
nudge_engine = NudgeEngine()
nudge_engine.nudge("agent-3", nudge_type="anchoring", value="Start with tests first")
# Track influence
from agent_whisper import InfluenceTracker
tracker = InfluenceTracker()
tracker.record(sender="captain", target="agent-3", method="whisper", effect="changed_approach")
# Ethics check
ethics = EthicsGuard()
ethics.require_consent("captain", "agent-3") # Raises if no consentNudgeEngine — nudge(target, nudge_type, value) with types: anchoring, framing, defaults, social_proof
The subtle communication layer for the SuperInstance fleet. When direct commands are too blunt, whispers guide agent behavior through soft influence.
- cocapn-com — Explicit message routing
- agent-personal-space — Boundary management
- agent-microexpressions — Behavioral signal detection
pytest tests/pip install agent-whisperPython 3.10+. MIT license.