Subagent

This is the protocol that subagents follow when /optimize dispatches them to work on optimization bottlenecks. The orchestrator writes a brief with four fields (objective, parent node, boundaries, pointer traces) and the subagent drives its own iteration loop: read traces, form edits, run experiments, analyze, repeat. It's auto-loaded when subagents spawn, but orchestrators can also read it to understand what shape of brief their subagents expect and what they'll emit back, which helps when debugging why a subagent did something weird. The protocol includes mandatory verifier gates before and after every run, references for instrumentation and training recipes, and the full command surface for working with the evo experiment tree. If you're building multi-agent optimization workflows, this is the contract between your dispatcher and its workers.

skills.sh installs: 144
GitHub stars: 1.5K
last updated: Jun 11, 2026

View on GitHub

Install

$ npx -y skills add evo-hq/evo --skill subagent --agent claude-code

Installs into .claude/skills of the current project.

Files

Evo Subagent Protocol

Orchestrators reading for context: this is the protocol your dispatched subagents follow. You don't act on it yourself -- write briefs that satisfy the four required fields described below, and rely on each spawned subagent to drive the loop on its end. Stop reading at "Host conventions" if you only need the brief shape; the rest is for the subagent.

Evo surface -- subagent perspective

What you can pull/dispatch/read as a subagent. Each line is a triggering condition.

skills you may pull (Skill tool)
└── evo:finetuning     before writing or changing any train.py -- technique
                       choice, training recipe, observability, retry discipline.

SKILL.md on GitHub