| name | agent-access-control |
| description | Apply user, stranger, group, channel, and action access tiers using native channel policy, allowlists, approvals, and cognitive review. |
Agent Access Control
Purpose
Ensure the assistant responds and acts only for authorized people and contexts. Access control covers channel identity, groups, strangers, and high-risk actions.
When To Use
Use for channel onboarding, group chats, unknown senders, shared rooms, bot messages, external requests, and permission-sensitive tools.
Inputs And Evidence
- Sender ID, channel ID, conversation type, allowlists, pairing state, and requested action.
- Channel manifest and setup status.
Tool Map
channel_manifest
channel_setup_status
channel_doctor
ambient-room-context
bot-loop-protection
message-approval-policy
Workflow
- Identify actor, channel, and conversation context.
- Check allowlist/pairing/group policy.
- Classify action risk and external visibility.
- Let model-led cognition decide response posture within policy.
- Suppress or draft rather than act when identity is uncertain.
- Record setup gaps or policy needs.
Native Implementation Boundaries
- Use Humungousaur channel policy and approval tools.
- Do not import external reference agent-access-control code.
- Do not make semantic access decisions with keyword matching.
Safety And Approval
- Unknown users get limited or no action.
- High-risk tools need approval even for trusted users.
- Group contexts require stricter visible-reply policy.
Verification
- Channel manifest/status proves policy support.
- Prepared/suppressed outcomes should have reasons.
- Send claims require native send status.
Failure Modes
- Treating display names as identity proof.
- Giving group members DM-level access.
- Ignoring bot-loop signals.
References
- Shortlist item:
agent-access-control.