[Bug] Colored Agent Badges Removed in System Prompt Optimization

Resolved 💬 3 comments Opened Oct 10, 2025 by eladmint Closed Oct 14, 2025

Bug Description
Colored agent badges removed in v2.0.11 system prompt optimization

Description

Colored agent badges that appeared when invoking subagents disappeared in v2.0.11. Through systematic version testing, I've identified the exact cause.

Root Cause

v2.0.11 CHANGELOG: "Reduced system prompt size by 1.4k tokens"

This optimization removed the badge rendering instructions from Claude's system prompt, eliminating the colored visual indicators for active agents.

Evidence

Systematic Version Testing Results

| Version | Date | Badges Working? |
|---------|------|----------------|
| v1.0.128 | - | ✅ YES |
| v2.0.9 | Oct 6 | ✅ YES |
| v2.0.10 | Oct 7 | ✅ YES |
| v2.0.11 | Oct 8 | ❌ NO (First broken) |
| v2.0.12 | Oct 9 | ❌ NO |
| v2.0.13 | Oct 9 | ❌ NO |

Exact breaking point: Between v2.0.10 (working) and v2.0.11 (broken)

What Badges Looked Like (v2.0.10 and earlier)

Working display:

  • Red colored badge/pill around agent name (e.g., debugger)
  • Agent name highlighted in assigned color
  • Visual distinction for active agent
  • Real-time status indicators

Current broken display (v2.0.11+):

  • Plain text with bullet point (●)
  • Agent name in parentheses
  • No colored badge
  • No visual distinction

Investigation Details

What We Ruled Out

  • ❌ Terminal capabilities (verified TERM=xterm-256color, COLORTERM=truecolor)
  • ❌ Agent configuration (color fields present in agent files)
  • ❌ Hooks/plugins (tested with all disabled)
  • ❌ Renderer rewrite (v2.0.10 renderer rewrite preserved badges)

What We Confirmed

  • ✅ System prompt reduction in v2.0.11 removed badge rendering
  • ✅ Exact version pinpointed through binary search testing
  • ✅ Issue is in Claude Code itself, not user configuration

Impact

Colored agent badges provided:

  1. Visual feedback for which agent is currently active
  2. Status awareness during multi-agent workflows
  3. Better UX for parallel agent execution
  4. Quick identification of agent context

Their removal reduces visibility into agent execution state, especially for users with multiple specialized agents.

Request

Could you please:

  1. Restore badge rendering in the system prompt (if token budget allows)
  2. Document if intentional - Was this removal intentional? Should users expect badges to return?
  3. Provide configuration option - Allow users to enable/disable badges based on preference
  4. Consider progressive enhancement - Make badges optional for users who want to save tokens

Workaround

Users can downgrade to v2.0.10 to restore colored badges:

npm install -g @anthropic-ai/claude-code@2.0.10

Trade-off: Miss bugfixes from v2.0.11+

Environment

  • Platform: macOS (Darwin 24.5.0)
  • Terminal: VSCode integrated terminal
  • Claude Code: v2.0.13 (tested v1.0.128 through v2.0.13)
  • Last working version: v2.0.10
  • First broken version: v2.0.11

Additional Context

This was discovered after noticing badges that were working on Sept 30 (v2.0.x early versions) disappeared after upgrading to v2.0.13 on Oct 9. Systematic testing across 6 versions identified v2.0.11's system prompt optimization as the exact cause.

The badge feature was particularly valuable for users managing complex multi-agent workflows with many specialized agents.

---

Related Issues:

  • #5259 - Feature request for colored subagent badges (requested Aug 2025)
  • #5254 - Related to subagent display (if applicable)

Environment Info

  • Platform: darwin
  • Terminal: vscode
  • Version: 2.0.13
  • Feedback ID: 8edef434-d4e0-4f7e-a12f-e37853605d12

Errors

[{"error":"Error: 1: 1 \n    at uL4 (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:559:3994)\n    at async hL4 (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:559:1604)","timestamp":"2025-10-10T17:50:33.969Z"},{"error":"Error: Language not supported while highlighting code, falling back to markdown: \n    at hK (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:2116:729)\n    at file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:2116:457\n    at Array.map (<anonymous>)\n    at tF (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:2116:448)\n    at UlB (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:2116:12745)\n    at G$ (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:73:19415)\n    at Sa (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:75:42814)\n    at S9 (file:///Users/eladm/.npm-global/lib/node_modules/@anthropic-ai/claude-code/cli.js:75:38355)\n    at Ym (file:///Users/eladm/.npm-g

Note: Error logs were truncated.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗