Subject index
Notes on trying models and building products with LLMs.
I spent a week building a tool that measures agent prompt changes, then found out it had been measuring nothing. Subagents inherit no instructions at all, and a fingerprint token in the spec is what proved it.
I built a tariff reclassification agent without access to duty rates. The frozen August evaluation reached 88.9% at eight digits on its DEAD_CODE stratum; this update separates that result from the six-digit baseline and the production triage path.
My June 2026 Duo CLI batch review workflow, updated October 8 with model identifiers, headless auto-approval, Credits billing, retryable completion markers, and timeout scripts.
My June 2026 Duo CLI review workflow, with an October 8 update covering versions, namespace access, model identifiers, Credits billing, and headless tool approvals.
I spent a week building a tool that measures agent prompt changes, then found out it had been measuring nothing. Subagents inherit no instructions at all, and a fingerprint token in the spec is what proved it.
I built a tariff reclassification agent without access to duty rates. The frozen August evaluation reached 88.9% at eight digits on its DEAD_CODE stratum; this update separates that result from the six-digit baseline and the production triage path.
My June 2026 Duo CLI batch review workflow, updated October 8 with model identifiers, headless auto-approval, Credits billing, retryable completion markers, and timeout scripts.
My June 2026 Duo CLI review workflow, with an October 8 update covering versions, namespace access, model identifiers, Credits billing, and headless tool approvals.
