Issue for this PR

Closes #

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

This change transforms this:

"experimental.chat.system.transform": async (input, output) => {
  console.log("session, system", input.sessionID, output.system);
},

into supporting this:

"experimental.chat.system.transform": async (input, output) => {
  if (input.agent === 'title') return output;
  console.log("session, agent, system", input.sessionID, input.agent, output.system);
},

How did you verify your code works?

Built on Arch Linux and modified DietrichGebert/ponytail like this:

diff --git a/.opencode/plugins/ponytail.mjs b/.opencode/plugins/ponytail.mjs
index 2fd3625..f8a0f1f 100644
--- a/.opencode/plugins/ponytail.mjs
+++ b/.opencode/plugins/ponytail.mjs
@@ -46,7 +46,13 @@ export default async ({ client } = {}) => {
     // Append the ruleset to the system prompt every turn.
     'experimental.chat.system.transform': async (_input, output) => {
       const mode = readMode();
-      if (mode === 'off') return;
+      // Avoid injecting ourselves into OpenCode title generation
+      if (mode === 'off' || _input.agent === 'title') return output;
       output.system.push(getPonytailInstructions(mode));
     },

OpenCode uses a title agent with a system message You are a title generator. You output ONLY a thread title. Nothing else. .... However, when a very small agent is used (like smollm2-135M-it) the additional system prompt injected by the ponytail plugin proves too much for the model to handle.

This change allows the plugin to filter out title agent requests and not inject system message for that puprpose — it is likely not intended to be used that way anyway.

Screenshots / recordings

If this is a UI change, please include a screenshot or recording.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

If you do not follow this template your PR will be automatically rejected.