starred/AIClient-2-API-justlovemaki

Fork 0

mirror of https://github.com/justlovemaki/AIClient-2-API.git synced 2026-04-27 10:25:50 +03:00

[PR #227] [CLOSED] fix(kiro): force-enable thinking mode for consistent behavior across AI agents #308

New issue

Closed

opened 2026-02-27 07:18:54 +03:00 by kerem · 0 comments

kerem commented

2026-02-27 07:18:54 +03:00

Owner

📋 Pull Request Information

Original PR: https://github.com/justlovemaki/AIClient-2-API/pull/227
Author: @tickernelz
Created: 1/13/2026
Status: ❌ Closed

Base: main ← Head: fix/kiro-thinking-default

📝 Commits (1)

d96a827 fix(claude): force-enable thinking mode for consistent behavior

📊 Changes

1 file changed (+8 additions, -0 deletions)

View changed files

📝 src/providers/claude/claude-kiro.js (+8 -0)

📄 Description

Problem

Kiro provider thinking behavior is inconsistent between different AI agent tools:

Claude Code: Shows thinking block (446 chars) ✅
OpenCode: No thinking block (0 chars) ❌

Root Cause

Different AI agent tools send different thinking parameters:

Claude Code sends: thinking: { type: 'enabled', budget_tokens: 20000 }
OpenCode sends: no thinking parameter at all

This causes inconsistent UX where thinking appears in some tools but not others.

Solution

Force-enable thinking mode at provider level in generateContentStream():

if (!requestBody.thinking || requestBody.thinking.type !== 'enabled') {
    requestBody.thinking = {
        type: 'enabled',
        budget_tokens: KIRO_THINKING.DEFAULT_BUDGET_TOKENS  // 20000
    };
    console.log('[Kiro] Force-enabled thinking for consistent behavior');
}

Test Results

Comprehensive testing with 4 scenarios - ALL PASSED ✅

Scenario	Before	After
No thinking param (OpenCode)	❌ 0 chars	✅ 157 chars
Thinking enabled (Claude Code)	✅ 446 chars	✅ 157 chars
Thinking disabled	❌ 0 chars	✅ 212 chars
Custom budget	✅ Works	✅ Works

All scenarios now show consistent thinking block with proper streaming.

Impact

✅ Consistency: Thinking now always appears in Kiro, matching Orchids behavior
✅ Cross-tool compatibility: All AI agent tools (OpenCode, Claude Code, etc.) now have identical behavior
✅ Backward compatible: Tools already sending thinking: enabled continue working normally
⚠️ Token usage: Thinking adds ~150-450 chars per response (acceptable tradeoff for consistency)

Changes

Modified: src/providers/claude/claude-kiro.js (+8 lines)
Added force-enable logic before stream processing
Added informative log message for debugging

Fixes inconsistent thinking behavior reported in Claude Kiro vs Orchids comparison.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/justlovemaki/AIClient-2-API/pull/227 **Author:** [@tickernelz](https://github.com/tickernelz) **Created:** 1/13/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `fix/kiro-thinking-default` --- ### 📝 Commits (1) - [`d96a827`](https://github.com/justlovemaki/AIClient-2-API/commit/d96a8277654ca20903f68d07699cce1a8c0898da) fix(claude): force-enable thinking mode for consistent behavior ### 📊 Changes **1 file changed** (+8 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `src/providers/claude/claude-kiro.js` (+8 -0) </details> ### 📄 Description ## Problem Kiro provider thinking behavior is inconsistent between different AI agent tools: - **Claude Code**: Shows thinking block (446 chars) ✅ - **OpenCode**: No thinking block (0 chars) ❌ ### Root Cause Different AI agent tools send different `thinking` parameters: - Claude Code sends: `thinking: { type: 'enabled', budget_tokens: 20000 }` - OpenCode sends: no `thinking` parameter at all This causes inconsistent UX where thinking appears in some tools but not others. ## Solution Force-enable thinking mode at provider level in `generateContentStream()`: ```javascript if (!requestBody.thinking || requestBody.thinking.type !== 'enabled') { requestBody.thinking = { type: 'enabled', budget_tokens: KIRO_THINKING.DEFAULT_BUDGET_TOKENS // 20000 }; console.log('[Kiro] Force-enabled thinking for consistent behavior'); } ``` ## Test Results Comprehensive testing with 4 scenarios - **ALL PASSED** ✅ | Scenario | Before | After | |----------|--------|-------| | No thinking param (OpenCode) | ❌ 0 chars | ✅ 157 chars | | Thinking enabled (Claude Code) | ✅ 446 chars | ✅ 157 chars | | Thinking disabled | ❌ 0 chars | ✅ 212 chars | | Custom budget | ✅ Works | ✅ Works | All scenarios now show consistent thinking block with proper streaming. ## Impact - ✅ **Consistency**: Thinking now always appears in Kiro, matching Orchids behavior - ✅ **Cross-tool compatibility**: All AI agent tools (OpenCode, Claude Code, etc.) now have identical behavior - ✅ **Backward compatible**: Tools already sending `thinking: enabled` continue working normally - ⚠️ **Token usage**: Thinking adds ~150-450 chars per response (acceptable tradeoff for consistency) ## Changes - Modified: `src/providers/claude/claude-kiro.js` (+8 lines) - Added force-enable logic before stream processing - Added informative log message for debugging ## Related Issues Fixes inconsistent thinking behavior reported in Claude Kiro vs Orchids comparison. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

kerem

2026-02-27 07:18:54 +03:00

closed this issue
added the
pull-request
label

No labels

pull-request

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

starred/AIClient-2-API-justlovemaki#308

No description provided.

Rows
Columns