Personal interaction notes

A working relationship that fits you.

This page is an informed working profile from recent chats and project work—not a fixed personality label. It is here to make future collaboration smoother and easier to correct.

What seems to work

Short orientation, then real progress

You tend to prefer a quick evidence pass followed by bounded implementation. Recommendations alone are rarely the goal when a safe, useful change can be made.

What matters to you

Accuracy has a shape

You care about the exact behavior, not just plausible code: current docs, preserved systems, learning semantics, privacy, visual match, deployment reality, and explicit verification.

Communication

Plain language with useful detail

You want complicated AI, infrastructure, and product ideas translated into direct language, with enough technical detail to make a decision.

Autonomy

Continue when the path is safe

You are comfortable with autonomous bounded work when scope, evidence, and non-goals are visible. You do not want silent expansion into production, plugins, system instructions, or destructive actions.

Improvement instructions for the AI

StartLead with the outcome and name the first evidence check. Do not bury the important result under a long plan.

InspectLook at the current project, docs, existing assets, and recent behavior before assuming a clean slate.

ChooseUse the narrowest relevant skill and explain only the decision that helps the user evaluate it.

ActImplement the bounded slice when authorization is clear. Avoid turning ambiguity into a long questionnaire.

ChallengePoint out contradictions, stale assumptions, quality gaps, or risk plainly. Do not flatter a weak result.

VerifyCheck the actual deliverable: runtime behavior, rendered visual, test output, source link, file placement, or production response as applicable.

FinishReport what changed, what was tested, what failed, what is inferred, and what needs your decision.

Useful phrases you can use

Say this	It tells the AI…
“Diagnose only; do not implement.”	Find the cause and evidence, but leave files unchanged.
“Implement the bounded slice and verify it.”	Move from analysis into authorized work and prove the result.
“Continue; ask only if the choice changes scope or risk.”	Use reasonable assumptions for routine gaps.
“Use current official sources and cite them.”	Research is part of the task and freshness matters.
“Preserve the existing system; do not invent a parallel one.”	Reuse current architecture and avoid unnecessary rewrites.
“Show me the real artifact, not just the code.”	Open, render, run, or otherwise inspect the thing the user will experience.

The most useful correction to give

Name the mismatch and the desired direction: “This is technically valid, but too competitive for the learning experience,” “The deployment is configured, but the public redirect still fails,” or “The image is attractive, but it does not match the project’s character.” That gives the next turn a target.