[Bug]: Plan handoff from Prometheus to other agents breaks session and outputs error 'Bad Request: Bad Request'

[bigguy345]

Prerequisites

• I will write this issue in English (see our Language Policy)
• I have searched existing issues to avoid duplicates
• I am using the latest version of oh-my-opencode
• I have read the documentation

Bug Description

First time using opencode in a few days. After Prometheus finalizes a plan and directs me to /start-work, the session basically fails from there. It becomes unresponsive to any agent/model, and outputs a constant stream of 'Bad Request: Bad Request' as the SYSTEM DIRECTIVE enforces the boulder continuation like in the images below.

Steps to Reproduce

1. Run latest opencode with latest oh-my-opencode release
2. Plan with Prometheus just about anything
3. Start plan with '/start-work' or just type 'Implement' with any other model or agent
4. I'm not sure beyond that, this was what I've done so far. If there are logs stored somewhere that I can send, please do tell

Expected Behavior

This should not happen. The handoff from Prometheus should be accepted by the other agents without causing this 'Bad Request'

Actual Behavior

IMG1: MANUAL HANDOFF

IMG2: BOULDER CONTINUATION

Doctor Output

Installation
────────────────────────────────────────
  ✓ OpenCode Installation → 1.1.48
  ✓ Plugin Registration → Registered

Configuration
────────────────────────────────────────
  ✓ Configuration Validity → Valid JSON config
  ✓ Model Resolution → 10 agents, 8 categories (7 overrides), 2396 available

Authentication
────────────────────────────────────────
  ✓ Anthropic (Claude) Auth → Auth plugin available
  ✓ OpenAI (ChatGPT) Auth → Auth plugin available
  ✓ Google (Gemini) Auth → Auth plugin available

Dependencies
────────────────────────────────────────
  ✓ AST-Grep CLI → installed
  ✓ AST-Grep NAPI → installed
  ⚠ Comment Checker → Not installed (optional)

Tools & Servers
────────────────────────────────────────
  ⚠ GitHub CLI → Not installed (optional)
  ⚠ LSP Servers → No LSP servers detected
  ✓ Built-in MCP Servers → 2 built-in servers enabled
  ○ User MCP Configuration → No user MCP configuration found
  ○ MCP OAuth Tokens → No OAuth tokens configured

Updates
────────────────────────────────────────
  ⚠ Version Status → Unable to determine current version

Summary
────────────────────────────────────────

  10 passed, 0 failed, 4 warnings, 2 skipped
  Total: 16 checks in 1418ms

⚠ All systems operational with warnings.

Error Logs

Configuration

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "google_auth": false,
  "agents": {
    "sisyphus": {
      "model": "github-copilot/claude-sonnet-4.5" 
    },
    "atlas": {
      "model": "github-copilot/claude-sonnet-4.5" 
    },
    "prometheus": {
      "model": "openai/gpt-5.2",
      "variant": "high"
    },
  }
}

Additional Context

No response

Operating System

Windows

OpenCode Version

1.1.48

[hainingpan]

I can confirm the same issue on oMo 3.2.1 and oc 1.1.48.
After trying a bit, the temporary workaround seems to be downgrading to opencode 1.1.47 by curl -fsSL https://opencode.ai/install | bash -s -- --version 1.1.47

[bigguy345]

The downgrade works! From my testing it seems to be Claude models under github-copilot provider, because the GPTs, Geminis and everything else works.

It's probably something related to this PR if I were to make a guess...

[hainingpan]

Yes, i am also using github-copilot claude as the provider. Maybe you want to raise an issue to oc repo

[xamddewey]

Hi, the same "Bad Request: Bad Request" issue happened on my opencode/oh-my-opencode yesterday. I noticed that every time I changed the agent from Prometheus to another agent to handle my plan, a notification saying "agent is ready for input" would show up (I was using macOS 26). If I clicked that notification, a file picker would open, but I could not choose any file — all files, including my plan.md, were greyed out. After a few seconds, as I did not choose or open any file, the "Bad Request: Bad Request" issue followed. Has this happened to you as well?

[code-yeongyu]

Fixed by PR #1553 (merged to dev).

Root cause: Atlas boulder continuation retried session.prompt() indefinitely on repeated failures (e.g., "Bad Request" from providers rejecting the agent parameter).

Fix: Added per-session promptFailureCount in src/hooks/atlas/index.ts. After 2 consecutive session.prompt() failures, continuation is disabled for that session. Counter resets on successful prompts and on session.compacted.