Why the coding agent ignores your rules.
It isn't just memory loss, and it isn't a bad prompt. Your most important rule is losing a competition for the model's attention, on Control Surfaces you can change and measure.
Olinave turns that into a measurable problem. The framework is a named taxonomy grounded in controlled experiments and Anthropic's own published guidance.
You can start with the free live Clinic.
One hour on why your coding agent ignores your CLAUDE.md — the controls, the warning signs, and one fix demoed live. Sign up on Luma.
From a diagnostic scan of your repo to a full day on your Claude Code setup.
Find out why your coding agent ignores you, learn how to fix it, or we fix it with you on your own repo.
The Claude Code Diagnostic Report
We scan your current Claude Code setup and hand you a scored map of your Control Surfaces, with every gap quoted straight from your own code in a full report.
Find out more → 02 / MasterclassAdvanced Claude Code Masterclass
live online · July 23A live instructor-led half day session covering the full framework. You'll see how the agent really works, where your rules quietly fail, and which rule belongs where.
Find out more → 03 / WorkshopClaude Code Diagnostic Workshop
A full day for your team: the framework in the morning, then a full afternoon strengthening your own repo with us. Includes a full diagnostic scan of your repo delivered before the workshop begins.
Find out more →How we work.
Suggestion vs instruction
Set suggestions apart from instructions. Most of what you write as a rule is a request; not a guarantee.
Gate it, or win attention
Every failure collapses to one choice: gate the rule, or make it win the attention competition.
Measure, don't trust
Output passing review is not evidence a rule was followed. We measure whether it held instead of trusting that it did.
Practitioners, not theorists.
Olinave is Philip Forshaw, ex-Apple AI Strategy & Operations lead, a technical leader with decades of training and workshop delivery behind him. The framework is patent-pending.
The method is a named taxonomy of the recurring failure modes, grounded in controlled experiments and Anthropic's own published guidance. Separately, we applied it to 300+ real Claude Code setups to measure how common each failure is.
Agent Harness Engineering — a named taxonomy grounded in controlled experiments and Anthropic's published guidance. Framework patent-pending.
Ready to see why your agent ignores you?
live online · July 23The Masterclass is a live instructor-led half day.