Tech Blog

In-depth technical writing from the harness9 core team, focused on architecture decisions, engineering implementation, and design trade-offs.

Articles

Article	Date	Tags
88.7% on the Full SWE-bench Lite 300: A harness9 Technical Report	2026-09-13	`harness9` `agent` `golang` `benchmark` `swe-bench`
Context Compaction Is Not Deleting History: How harness9 Keeps an Agent on Track	2026-08-11	`harness9` `agent` `golang` `context` `compaction` `memory`
Benchmarking Beyond Scores: How harness9 Iterates from Execution Traces	2026-07-24	`harness9` `agent` `golang` `benchmark` `swe-bench` `terminal-bench`
Observability: Giving the Agent a Telescope Into Its Own Machinery	2026-07-20	`harness9` `agent` `golang` `observability` `opentelemetry` `langfuse`
Sandbox: Locking the Agent's Hands and Feet on a Floating Island	2026-07-13	`harness9` `agent` `golang` `sandbox` `docker` `isolation`
Sub-Agent: How harness9 Lets the Main Agent Delegate Work	2026-07-06	`harness9` `agent` `golang` `SubAgent` `task-delegation`
Hooks and Human-in-the-Loop: harness9's Tool Permission Interception System	2026-06-29	`harness9` `agent` `golang` `hooks` `permission` `human-in-the-loop`
Context Engineering: How an Agent Stays Sane Within a Limited Token Window	2026-06-22	`harness9` `agent` `golang` `context-engineering` `memory` `compaction`
The Agent Skills System: Extending LLM Capabilities via Progressive Disclosure	2026-06-15	`harness9` `agent` `golang` `skills` `progressive-disclosure` `prompt-engineering`
The Planning Module: Plan Mode, TodoStore, and Execution Automation	2026-06-08	`harness9` `agent` `golang` `planning` `todo` `stagnation-detection`
harness9's Tool Calling System: From Interface Contracts to Concurrent Sandboxing	2026-06-01	`harness9` `harness` `agent` `golang` `tool-calling` `concurrency` `sandbox`
Agent Loop: A Production-Grade ReAct Loop in 500 Lines of Go	2026-05-28	`harness9` `agent` `golang` `react` `concurrency`