A Harness Engineering Best Practices case completed in collaboration with Codex. Today, the friends saw that Codex was about to reset the quota again, so they adjusted the model to 5.5 high and grasped the token at N times the speed. At the same time, I opened the gpt 5.5 high window at 8, because I have been playing goal more and more recently, so the whole process is still pretty cool. However, I still encountered a number of tricky problems when performing complex tasks. The biggest headache is that because the initial goal itself is complex, it will take more than 2 hours to complete a series of workflow designs, similar to a small agent. The biggest challenge encountered is that it will compress my goal into a number of more end-to-end local goals. The advantage of this is that the model is relatively clear to process, and the context will not be messed up. The disadvantage is the content of my screenshots. It always unconsciously associates these small goals. Since the large language model naturally does not have a stable task execution boundary, it will perform the next step that seems reasonable, so there will be a gap in it. This is also why we do harness engineering optimization, context optimization, memory... all to reduce the cost of friction between the model and the environment, the user. Codex also admitted some of its current design shortcomings, so I created a harness skill to reduce model and environmental friction, debugging for a few days first, and open source to everyone if it is easy to use. What Codex itself has not yet done, the most appropriate mechanism to supplement: - Atomization execution: Make "Generate + Publish + Validate" an indivisible default action. - Mandatory Validation: Failure to pass Publish Validation, Format Validation, Link Validation is not complete. - Explicit task mode: Initialization and validation entries are published by default, making it a harness that can be reused across tasks.
Visual summary
Article argument map
Generated from the post's content graph
Mermaid outline
flowchart LR
format-long_post["longform note"]
topic-agents["agents"]
topic-harness-engineering["harness engineering"]
topic-ai-workbenches["ai workbenches"]
topic-memory["memory"]
capability-agent-workflow["agent workflow"]
capability-harness-engineering["harness engineering"]
capability-ai-native-workbench["AI-native workbench"]
capability-evaluation["evaluation"]
market-open-source-builders["open-source builders"]
format-long_post -->|covers| topic-agents
format-long_post -->|covers| topic-harness-engineering
format-long_post -->|covers| topic-ai-workbenches
format-long_post -->|covers| topic-memory
format-long_post -->|signals| capability-agent-workflow
format-long_post -->|signals| capability-harness-engineering
format-long_post -->|signals| capability-ai-native-workbench
format-long_post -->|signals| capability-evaluationVisual structure
Essay structure map
Built from summary and key paragraph positions
Mermaid outline
flowchart LR
thesis["A Harness Engineering Best Practices case completed in collaboration with Codex. Today, the friends saw tha..."]
signal["A Harness Engineering Best Practices case completed in collaboration with Codex. Today, the friends saw tha..."]
operator["A Harness Engineering Best Practices case completed in collaboration with Codex. Today, the friends saw tha..."]
implication["A Harness Engineering Best Practices case completed in collaboration with Codex. Today, the friends saw tha..."]
thesis -->|frames| signal
signal -->|develops| operator
operator -->|lands in| implication