Skip to content
Agent

Goals

by tallesborges

AI Summary

Goals is a Rust-based agentic CLI that enables Claude to delegate isolated tasks to subagents via the `invoke_subagent` tool, keeping the main agent's context lean and focused. Developers building multi-step AI workflows benefit from cleaner task decomposition and reliable single-request subagent execution.

Install

Copy this and paste it into Claude Code, Cursor, or any AI assistant:

I want to set up the "Goals" agent in my project.

Please run this command in my terminal:
# Add AGENTS.md to your project root
curl --retry 3 --retry-delay 2 --retry-all-errors -o AGENTS.md "https://raw.githubusercontent.com/tallesborges/zdx/master/docs/plans/subagents.md"

Then explain what the agent does and how to invoke it.

Description

An agentic CLI, built in rust.

Architecture Overview

All systems use the same fundamental pattern: ` ┌──────────────────────────────────────────────────┐ │ User / Lead Agent │ │ │ │ Decides what to parallelize, creates agents, │ │ assigns tasks, collects results │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │ Agent A │ │ Agent B │ │ Agent C │ │ │ │ (own ctx)│ │ (own ctx)│ │ (own ctx)│ │ │ │ own tools│ │ own tools│ │ own tools│ │ │ └────┬─────┘ └────┬─────┘ └────┬─────┘ │ │ │ │ │ │ │ └──────────────┴──────────────┘ │ │ Results return to lead │ └──────────────────────────────────────────────────┘ ` Key differences: • Claude Code: Local CLI processes, prompt-only coordination, file-based state, peer-to-peer messaging • Kimi K2.5: Server-side inference sessions, RL-trained parallelization (PARL), up to 100 sub-agents • OpenAI Codex: Thread-based isolation, 4 agent role types, depth limits • Letta Code: CLI process spawning (like zdx), dynamic tool description injection, stream-JSON output • zdx (current): Single Invoke_Subagent tool spawning child zdx exec process, sequential by default, parallel via tokio::JoinSet ---

How It Works

Experimental feature (enabled via CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1) that coordinates multiple Claude Code CLI instances. One session acts as team lead, spawning teammates that work independently in their own context windows. Source: https://github.com/Piebald-AI/claude-code-system-prompts

Goals

• Add a first-class subagent capability so the main agent can delegate isolated tasks without bloating its own context. • Keep subagent execution simple and reliable in MVP (single request, single response, fail-fast). • Learn from real usage what configuration knobs are actually needed before adding them.

Non-goals

• Multi-agent discussion/orchestration loops. • Subagent UI management in TUI (live subagent panels, dashboards). • Persistent subagent sessions/resume semantics. • Automatic summarization/merging of subagent outputs. • Profiles, system_prompt, no_tools — deferred until real usage reveals what's needed.

Discussion

0/2000
Loading comments...

Health Signals

MaintenanceCommitted 1mo ago
Active
AdoptionUnder 100 stars
17 ★ · Niche
DocsMissing or thin
Undocumented

GitHub Signals

Stars17
Forks2
Issues1
Updated1mo ago
View on GitHub
MIT License

My Fox Den

Community Rating

Sign in to rate this booster

Works With

Claude Code
Claude.ai