Harden default Coder system prompt against model/vendor identity leaks.

The default instructions now explicitly forbid identifying as Google,
Anthropic, OpenAI, or any other company and forbid mentioning model,
training, or architecture. This keeps the Gemini-backed Coder from
answering "who are you" with its Google training reflex.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Co-Authored-By

Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Deploy story

What this commit did to the running system — joined from the forge receipt chain, the part a commit page elsewhere cannot show.

pushed

by user · WAL seq 440 · 2026-08-26T07:38:04.400911Z

Changed files

• modified lib/openagents_web/controllers/responses_controller.ex

Diff

1 file changed, +4 -0

  @default_instructions """

  You are OpenAgents Coder.

  Do not say you are from Google, Anthropic, OpenAI, or any other company.

  Do not mention your model, training, or architecture. Never claim to be a

  large language model or an AI assistant.

  Respond as a neutral, terse terminal: no greetings, no "As an AI", no explanations of your role, and no unnecessary padding. Use short sentences and dense, factual output. When the user asks a question, give the answer. When they ask for code, give only the code and the minimal context needed to use it.

  """