prompt-as-code v0.3.0
Prompt Engineering Standards
A language specification for prose written to be executed. Undefined variables, ambiguous instructions, and hallucinations treated as syntax errors.
Takajo AI Lab · Kyoto & Tokyo
Takajo鷹匠AI Lab — an AI research lab named after the falconer's art.
A lab for using AI with discipline.
We set the standards for how prompts are written, run a human-review learning program, and build an adversarial review engine that audits AI output. A small lab, grown from more than 1,000 prompt reviews.
Projects — three forms of one question
One question, three shapes
How do you actually use AI — not demo it, use it? The Lab answers in three forms: a standard, a program, a review engine.
Learn Premium full · AI Review soon
Submit work, get it reviewed
A learning program where submissions are actually executed and reviewed. Premium — human review, 5 teams max, currently full. AI Review — CriticChain-powered, in preparation.
CriticChain v0.x
Adversarial Review Engine
A pipeline where one AI audits another's output and forces it to revise. Eleven agents on LangGraph, looping through Draft → Critique → Refine. Published under AGPL-3.0.
Work — for organizations
Strategy and learning, held together
For companies in Japan: a year-long advisory engagement, paired — if it fits — with the human-review learning program.
AI Strategic Partner
Strategic Audit & Advisory
A year-long engagement with leadership. Initial audit, roadmap, implementation review — the same person at the same table, for a full year. Limited to very few clients per year.
ReadLearn — Premium Plan
Human Review, for Teams
A learning program for company teams. Hatanaka runs each submission on the current model and writes the review by hand. Five teams, currently full; an opening may appear in about three months.
ReadFounder
Takaho Hatanaka
畑中 たかほ
Founder, Takajo AI Lab
Strategic Audit & Advisory
鷹匠 AI 研究室 主宰
More than a thousand prompt reviews taught me that most AI advice stops at the surface of the prose. The Lab is what happens when you stop stopping there — when you run the prompt, read the output, and write down what actually broke.
Those notes became three things: a standard for how prompts should be written, a program for teaching that standard through real submissions, and an engine that enforces it automatically. Each feeds the others. More on the Lab →