krafton-ai / KIRA

A smarter agent harness for Terminal-Bench , built on Terminus 2 Simple fixes, significant gains. What is Terminus-KIRA? Terminus-KIRA is an agent harness for Terminal-Bench, built on top of Terminus 2. It boosts frontier model performance on Terminal-Bench through a set of minimal but effective harness-level improvements — native tool calling, multimodal support, execution optimization, and smarter completion verification. --- Key Features • **Native Tool Calling** — Replaces ICL JSON/XML parsing with the LLM parameter for structured, reliable outputs • **Image Analysis (Multimodal)** — tool for base64-encoded image analysis directly from the terminal • **Marker-based Polling** — Early command completion detection using echo markers, cutting unnecessary wait time • **Smart Completion Verification** — Double-confirmation checklist covering requirements, robustness, and multi-perspective QA (test engineer, QA engineer, user) • **Prompt Caching** — Anthropic ephemeral caching on recent messages to reduce latency and cost --- Architecture Terminus-KIRA extends Terminus 2 by replacing its ICL (In-Context Learning) response parsing with native LLM tool calling. **Tool definitions** passed via the parameter: | Tool | Purpose | |---|---| | | Run shell commands with analysis and plan | | | Signal task completion (triggers double-confirmation) | | | Analyze image files via base64 multimodal input | **How it works:** • Calls directly with , bypassing the base class to access native tool calling • The model returns structured tool calls instead of free-form text — no regex/JSON parsing needed • On context window overflow, automatically summarizes conversation history and retries • Marker-based polling appends after each command; if the marker appears before the requested duration, execution moves on immediately --- Evolution Key milestones from development history: | # | Milestone | Description | |---|---|---| | 1 | Genesis | Copy of Terminus 2 as starting point | | 2 | Native Tool Use | Replaced ICL JSON/XML parsing with LLM parameter | | 3 | Output Limiting | 30 KB cap on terminal output to prevent context bloat | | 4 | Autonomy & Constraints | Prompt engineering for agent autonomy and environment constraints | | 5 | Completion Confirmation | Include original instruction in completion check | | 6 | Multimodal | tool for visual analysis of terminal screenshots | | 7 | Completion Checklist | Multi-perspective QA checklist (test engineer, QA, user) | | 8 | Execution Optimization | Marker-based polling and block timeout protection | | 9 | Temperature Fix | Set temperature to 1 when using reasoning effort | --- Usage For more details, visit our blog post. --- Project Structure --- Citing Us If you found Terminus-KIRA useful, please cite us as: --- Changelog | Version | Description | |---|---| | **v1.1** | Migrated from In-Context Learning (ICL) to **native tool calling** via LLM parameter. Removed verbose JSON/XML response format instructions from system prompt — the model now receives structured tool definitions directly, resulting in a significantly shorter prompt and more reliable outputs. | | **v1.0** | Initial release. Fork of Terminus 2 with ICL-based JSON response parsing and full response format instructions in the system prompt. | --- KRAFTON AI & Ludo Robotics