bot-loop-protection

name	bot-loop-protection
description	Prevent bot-to-bot reply loops by detecting bot-authored inbound messages, pair suppression, channel policy, and ambient context rules. Use when handling channels that may receive messages from other assistants or automation.

Bot Loop Protection

Purpose

Keep assistants from endlessly replying to each other. This skill adapts external reference bot-loop protection concepts to Humungousaur's native channel metadata and policy gates.

When To Use

Use for Slack, Discord, Telegram, Teams, WebChat, SMS, or any channel where inbound messages may be bot-authored or automation-authored.

Inputs And Evidence

• Inbound metadata: bot_authored, author type, sender ID, message ID, conversation ID, and channel ID.
• Channel manifest loop-protection support.
• Recent messages and outbox entries.
• Group/DM policy.

Tool Map

• channel_manifest
• channel_catalog
• channel_outbox
• activity_ingest
• cognitive_interaction_review
• channel_message_prepare

Workflow

1. Inspect inbound metadata before considering message content.
2. If the message is bot-authored and the channel does not allow bot-authored inbound handling, suppress response.
3. For supported bot-authored handling, check pair/session policy and recent outbox to avoid ping-pong.
4. Convert suppressed messages into passive context only when useful and safe.
5. Let model-led cognition review ambiguous cases; do not build hardcoded phrase triggers.
6. Record suppression reason for diagnostics.

Safety

• Never let two bot accounts repeatedly call each other's message tools.
• Do not bypass bot-loop protection because the text appears urgent.
• Do not import external reference bot-loop code directly.

Native Implementation Boundaries

• Use native channel metadata, outbox history, and cognitive review to distinguish bot-authored context from user intent.
• Use activity_ingest for safe passive context and channel_message_prepare only when policy allows a response.
• Treat missing bot metadata from an adapter as a channel implementation gap; do not infer bot identity from vague text patterns.

Verification

• Suppressed events should show a reason such as bot-authored or pair suppression.
• Prepared replies should not target bot-authored messages unless policy explicitly allows it.
• Outbox should not show repeated alternating bot replies.

Failure Modes

• Missing bot identity metadata from a channel adapter.
• Treating assistant-generated text as user intent.
• Suppressing all automation messages even when the user configured them as signals.

References

• Shortlist item: bot-loop-protection.
• Upstream inspiration: external reference bot-loop protection.