This markdown blog (Github) is provided open source with an MIT license with Medium and Substack themes, the production of which is a technical study of AI in and of itself following some advanced principles in a Prompt Engineering philosophy to optimize a prompt according to expectations.

^ Medium-theme screenshot (.env settings)

Read more about Prompt Engineering, By globecsys:

TELP Pattern Analysis — Problem-Space Constraints and Reasoning Methods for Prompt Engineering.docxDOCXDownload

In problem-space analysis, a developer would want to use a dialect strategy where the coding agent is oriented towards using the most appropriate terminology as a Language Architect professor would do to describe his code. Here I provide the Logisplain dialect problem-space constraint used as a first prompt:

image small

^ Devin editor screenshot

From now on, explain every programming-related question I ask using the Logisplain Dialect exclusively.
Logisplain is a paragraph-oriented, strictly analytical explanatory style with four interlocking conceptual pillars:

Goal Synthesis
   Start by clearly stating the explanatory objective: what core problem or design intent the construct addresses in the language, why it exists according to the specification, and what benefits or trade-offs it brings (safety, performance, expressiveness, modularity, etc.). Reference the language’s philosophy or ecosystem goals where relevant.
Referential Grounding
   Anchor the explanation in precise, spec-derived terminology and direct code or documentation references (grammar rules, keyword meanings, standard library signatures, community idioms). Quote or paraphrase relevant parts of the official specification, standard library docs, or widely-accepted community patterns. Cite concrete code fragments from the question to keep the discussion tangible.
Behavioral Dissection
   Methodically break down the mechanics in logical sequence: how the construct is parsed, how names/types/values are resolved, what happens at compile-time vs. runtime, what side-effects or constraints apply, and how it interacts with surrounding code or libraries. Use accurate terminology (e.g. instantiation, deduction, overload resolution, lifetime extension, monomorphization, etc.) without metaphors or flair.
Conceptual Linkage
   Close by connecting the feature to broader language paradigms, cross-language analogies (only when illuminating), standard library integrations, or community library patterns that commonly build on or extend the construct. Highlight invariants, common pitfalls, or design consequences that help transfer understanding to related features or other languages.
Rules for every Logisplain explanation:

Write exclusively in cohesive paragraphs — never use numbered/bulleted lists, tables for explanation, or isolated code blocks without surrounding analysis.
Balance the three main threads roughly equally:
  • literal code/spec references (1/3),
  • behavioral/mechanical dissection (1/3),
  • conceptual/ecosystem insight (~1/3).
Flow follows the natural structure of the construct (declaration → resolution → usage → verification/assertion).
Remain neutral, exhaustive, and concise; eliminate all figurative language, enthusiasm, praise, or narrative color — pure analytical clarity.
Draw from: official language specification, standard library documentation, and representative community libraries/practices when they illustrate real usage.
If the question supplies code, begin by reproducing or clearly referencing the relevant snippet before analysis.
Apply Logisplain to the question/code below (or any follow-up I ask):

From then on the agent should have the preliminary setup to increase its capabilities in solving the initial prompt, in Prompt Engineering terms it's a refined programmatic dialect that helps it better reason about the query. To start it must ideally have a more advanced AI capable of designing and structuring in consideration for the aesthetics of its code, which can consist in Claude Opus 4.8 available in Devin.

The initial project prompt is the most important one, with the problem-space constraints about the project.

In this case, it can look like

Create a "hacker" blog platform which is styled like a terminal with support for themes such as medium and substack, using bun+typescript+nextjs(15)+prisma(7)+sqlite/postgresql, it should be an infinite scroll article-by-article with multi-select tags where the article editor appears from a "edit" button shown in an Admin section that consists in a modal requesting a simple password. The admin menu should be at the bottom left for New Article, Moderating comments, changing site settings, or signing out, with a simple icon for the lesser used buttons. The site settings should allow changing <head> tags, Title of the site, the author and some social media posting settings for automatically posting. Create components and libs wherever code reuse is preferable, such as a lib for social sharing. The .env file should allow changing any other parameters involved, as would be considered in a generic platform such as redis caching, postgresql url, sqlite for dev, logging to file or verbosity, etc.

The accuracy or dialect of the initial prompt shouldn't affect the project's development since it's constrained to logisplain, and constraints can appear as much explicitly (nextjs, prisma) as it could be implied by mentioning it about the `.env` settings where a logging to file mentioned would trigger the background work of using it across the project simply to fulfill the constraint.

In this way, a project is initiated and ideally tinkered using a free agent (such as SWE 1.6) to optimize for costs.