The Ultimate Guide to /goal

2026-04-29 · Source: unwind ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

The article introduces "/goal" as a new primitive for coding agents, akin to HTTP or JSON, enabling agents to autonomously work towards a defined "done" state. Unlike traditional prompts that require constant human steering, /goal allows users to submit a task once and have the agent pursue it until completion or budget exhaustion. OpenAI's Codex CLI and Anthropic's Claude Code have recently adopted /goal, alongside the Hermes Agent orchestrator. Hermes Agent, running on a Mac Mini, coordinates tasks between these coding workers, transforming individual /goal commands into a managed workflow displayed on a Kanban board. This system allows for automated setup, task decomposition, worker selection, and crucial post-completion verification, ensuring that agents' self-reports are validated against actual test and build outcomes.

Key takeaway

For AI Architects designing autonomous agent systems, adopting the /goal primitive is critical for enabling robust, verifiable workflows. Your systems should integrate /goal for task assignment and implement independent verification steps to validate agent outputs, transforming agent interactions from reactive prompting to proactive, contract-based execution. This approach allows for scalable, parallel development and review processes, reducing manual oversight and increasing reliability.

Key insights

/goal shifts agent interaction from turn-by-turn prompting to autonomous, goal-driven task assignment with defined completion criteria.

Principles

• Define "done" explicitly for agent tasks.
• Verify agent output; do not trust self-reports.
• Orchestrators enable complex agent workflows.

Method

Submit a /goal command specifying desired outcomes (e.g., "tests pass, build passes"). An orchestrator like Hermes then decomposes the task, assigns it to suitable coding workers (e.g., Codex for building, Claude Code for reviewing), and verifies completion.

In practice

• Use /goal for hands-off agent task execution.
• Implement a verifier for agent-generated code.
• Employ an orchestrator for multi-agent workflows.

Topics

• /goal Primitive
• Coding Agents
• Agent Orchestration
• Automated Verification
• Software Development Workflow

Best for: AI Architect, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• How to Use /Goal to Do More With AI — /goal transforms AI interaction from turn-based prompts to autonomous, self-evaluating, goal-driven loops.
• Autonomous Long-Running Coding Agents — Autonomous coding agents require robust control systems with goals, evaluators, verifiers, and loops for long-running, verifiable task completion.
• Critique of Agent Model — Genuine AI agency requires internalizing goal, identity, decision-making, self-regulation, and learning structures, not external scaffolding.
• Claude Code's '/goals' separates the agent that works from the one that decides it's done — Separating task execution from evaluation prevents premature AI agent exits and improves reliability.
• Your First AI Agent — How to Build Autonomous Workflows That Work While You Sleep — Prompt to… — AI agents autonomously pursue multi-step goals, making decisions and using tools, unlike single-query prompts.
• Codex Goals enable the model to work autonomously for hours without supervision — AI goals enable autonomous, long-term problem-solving for code quality issues without constant supervision.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by unwind ai.