XDOG

Tools for building AI agents and managing LLM deployments.

GitHub →

xdog / Packages / skills / Features

Common goal/plan instructions and isolated Goal domain support, not an Agent harness.

Instructions

Discoverable skills
goal and plan are listed without injecting full bodies by default; Flow/Eyes ship their own skills.

Goal

Native Goal commands
Installed Coding/Claw capability contributes /goal commands and model tools independently of activation.
Agent-owned workflow
The current main Agent plans, delegates optionally, evaluates progress and judges completion.
Authoritative files
One current Goal per owner; task JSON, plan.md, progress.md and versioned registry history.
Progress reports
Before each continuation the main Agent records actual progress; repeated no progress can lead it to pause.

Integration

Signal-driven reminders
Generic host events and optional Eyes task observation coalesce into ACP reminders to the original main Agent.
Explicit lifecycle
Pause stops proactive reminders/owned observation; completed remains visible until clear; no host wait or timer auto-authorization.