From f830c0cb23741ced83011fcc9825c743554cd7c7 Mon Sep 17 00:00:00 2001
From: wassname <1103714+wassname@users.noreply.github.com>
Date: Wed, 19 Aug 2026 20:55:43 +0800
Subject: [PATCH] skill: put literal trigger phrases in the description so it
 self-invokes

A sonnet subagent had ml-debug listed, with 'after a run finishes or crashes' in
the description, and did not invoke it for 'read the last pueue log'.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
---
 SKILL.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/SKILL.md b/SKILL.md
index 8efb27b..1ad0441 100644
--- a/SKILL.md
+++ b/SKILL.md
@@ -1,6 +1,6 @@
 ---
 name: ml-debug
-description: "Procedure for machine learning development and debugging. Execute the numbered steps and paste the filled templates into your reply. Use before queueing a run, after a run finishes or crashes, when stuck, before reporting any result, and before writing that a result looks fine. Makes 'read your data', 'assume you have a bug' and 'compare to reference code' produce artifacts the user can check."
+description: "Procedure for machine learning development and debugging. Execute the numbered steps and paste the filled templates into your reply. Invoke it yourself, do not wait to be asked. Triggers: read the log, read the last pueue job, check the run, what did that job show, the training run finished, it crashed, queue a run, the loss is not going down, the metric will not move, is this result real, does A beat B, and any moment you are about to write that a result looks fine. Makes 'read your data', 'assume you have a bug' and 'compare to reference code' produce artifacts the user can check."
 ---
 
 # ML procedure: execute, do not summarise