Agents & Channels
This page explains each published agent’s permission profile and why it is isolated that way.
The public bundle includes agent identity and a simplified permission/config posture. Exact peer bindings, raw filesystem paths, and detailed private allowlists are still omitted.
Models
- Primary model:
github-copilot/claude-sonnet-4.6 - Fallback models:
github-copilot/gpt-5.4 - Primary image model:
github-copilot/claude-sonnet-4.6
Agent Architecture
Each published agent has its own permission boundary. Interactive helpers stay separated from delegated workers and webhook-driven automations.
| Agent | Used for | Permissions | Why it is set up this way |
|---|---|---|---|
main | Jeff’s primary direct chats and proactive assistant flows | customized tools; exec allowed; browser default; writes default; sub-agents root, family, finance. | Keeps the everyday assistant capable with exec restricted to safebin CLIs only — no arbitrary shell access. |
mail | Internal delegated mail-processing workflows | profile:minimal tools; read allowed; writes denied; browser denied; exec denied. | Treats mail as untrusted input and isolates mail processing from broader tools. |
root | Explicit owner escalations for admin/debugging work | inherited-default tools; broad inherited access posture; exec default. | Concentrates privileged admin/debug access in a separate escalation path. |
family | Family-facing direct chats | profile:messaging tools; writes denied; browser denied; exec denied; sub-agents none. | Limits family-facing conversations to a narrow, safer tool surface. |
finance | Published agent surface | profile:minimal tools. | Separates this agent from the rest of the system. |
hass-hooks | Home Assistant webhook events | custom-allowlist tools; tightly scoped allowlist for camera, image, and message handling only. | Ensures webhook automation can inspect camera events and notify, but not wander outside that workflow. |
coding | Coding specialist in #coding on Discord | inherited-default tools. | Dedicated coding agent for code review, architecture, ACP agent delegation, and infra/DevOps work. Kept separate from main to allow elevated shell access without exposing it to everyday chat. |
notify | Published agent surface | custom-allowlist tools. | Separates this agent from the rest of the system. |
Exec & Safebin
Exec permissions are configured per agent using agents.list[].tools.exec:
main—security: allowlist— can only run binaries listed insafeBins(vetted CLI scripts in~/safebin/)root—security: full— unrestricted shell access for admin/debug escalation- Other agents — exec denied entirely
See CLI Tools for the available safebin inventory.
Channels
| Channel | Enabled | DM Policy | Group Policy | Streaming |
|---|---|---|---|---|
discord | Yes | pairing | allowlist | partial |
telegram | Yes | pairing | allowlist | off |
Session Settings
| Setting | Value |
|---|---|
| Scope | per-channel-peer |
| Reset mode | idle |
| Reset hour | 4 |