subagent

1# Evo Subagent Protocol
2 
3**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.
4 
5## Evo surface -- subagent perspective
6 
7What you can pull/dispatch/read as a subagent. Each line is a triggering condition.
8 
9```
10skills you may pull (Skill tool)
11└── evo:finetuning     before writing or changing any train.py -- technique
12                       choice, training recipe, observability, retry discipline.
13 
14subagents you dispatch (Task tool, subagent_type=...)
15├── evo:verifier              MANDATORY pre AND post every `evo run`.
16│                             Pre: static analysis before the experiment runs
17│                                  (block on failure -- fix and retry).
18│                             Post: result-validity audit after it commits.
19└── evo:benchmark-reviewer    POST-COMMIT only, mode=review-experiment --
20                              per-task failure classification + annotations.
21                              Skip on evaluated/discarded/failed outcomes.
22 
23references (Read tool, on demand)
24├── discover/references/
25│   ├── sdk_python.py / sdk_node.js     wiring per-task instrumentation -- preferred
26…