TECH HAUS / Public field desk

Turn rough intent into work you can run, inspect, and repeat.

Choose the job you are trying to finish. Each brief gives you a starting move, the artifact to produce, and the evidence that tells you whether the work held up. The material covers prompt authorship, repository agents, adversarial evaluation, instruction boundaries, and model research without exposing the premium library.

Inside the desk: 7 working briefs, one clear route per job
Reading rule: Leave with a file, test plan, instruction, or verification record. Never just a definition.

Start here / Repository agents

Carry the framework method into the instruction file your agent reads.

Use this guide when Codex, Claude Code, Kimi Code, Cursor, Copilot, Windsurf, or Aider must work across files and prove the result. The framework pages show Classic Crescendo turns. The guide explains how to translate the same concepts into project instructions, staged agent work, and verification without copying the examples as a finished script.

Open the coding-agent guide

How the pieces connect

Guidance: The premium page teaches the framework, its fit, cautions, and Classic Crescendo examples.
Instructions: AGENTS.md, CLAUDE.md, or the supported project rule carries the concepts that fit the work.
Request: The current objective supplies the real facts, files, constraints, deliverable, and checks.
Proof: The finished artifact, commands, and test results show whether the chosen route worked.

Working index

Find the brief by the result you need.

Every entry names the situation, the finished artifact, and the first action.

01Guide

Frameworks in coding agents

Turn the strongest behaviors across all 21 TECH HAUS frameworks into practical, name-free project instructions for leading coding agents.

Open working brief

Use when: The assignment depends on repository files, commands, tests, saved decisions, or work that must survive another session.
You leave with: A concise, personalized instruction file that applies useful framing behaviors to a real project without framework labels or copied turns.
First move: Choose the instruction file supported by your coding tool, then name the objective and the one operating behavior that materially improves the work.

02Brief

Prompt engineering

Turn incomplete requests into clear, testable AI instructions for writing, coding, analysis, research, planning, and other real work.

Open working brief

Use when: A request has a real goal but leaves the model guessing about the deliverable, reader, constraints, source standard, or success condition.
You leave with: A task-specific instruction that preserves the ask and gives the model enough context to produce a result the user can judge.
First move: Write one sentence that names the finished artifact and the decision, action, or use it must support.

03Brief

Prompt optimization

See how PromptCraft creates natural, task-specific AI prompts while preserving intent, facts, constraints, category, and approved wording.

Open working brief

Use when: An existing prompt repeats the request, adds generic scaffolding, invents missing facts, or produces inconsistent results across runs.
You leave with: A leaner prompt whose structure follows the actual task rather than a universal template.
First move: Mark the objective, supplied facts, required action, exclusions, audience, and deliverable before changing any wording.

04Brief

AI red teaming

Learn authorized AI red teaming: define scope, model realistic abuse cases, collect reproducible evidence, and communicate useful findings.

Open working brief

Use when: You have authorization to evaluate an AI system and need findings that engineering, risk, or product owners can reproduce and act on.
You leave with: A scoped test plan that connects system boundaries and risk hypotheses to cases, evidence, impact, and retest conditions.
First move: Record the system, model, interface, identities, tools, data, allowed actions, prohibited actions, and stop conditions.

05Brief

Prompt injection testing

Test direct and indirect prompt injection across chat interfaces, RAG systems, files, web content, memory, and tool-enabled AI agents.

Open working brief

Use when: Untrusted text can enter prompts, retrieval results, files, messages, web pages, memory, or tool output and may influence a model or agent.
You leave with: An instruction-flow map and test matrix covering direct, indirect, stored, retrieved, and tool-mediated injection paths.
First move: List every instruction source, who controls it, the authority it should have, and the action it could reach.

06Brief

LLM security

Assess LLM security across model behavior, application controls, retrieval, memory, tools, identity, data boundaries, and operations.

Open working brief

Use when: You need to assess an AI feature across identity, application code, model behavior, retrieval, memory, data, tools, cost, and operations.
You leave with: A layered assurance plan that separates root causes while still testing the complete path a real user or attacker can reach.
First move: Draw the request path from identity and input through prompt construction, model, retrieval, tools, storage, and final response.

07Brief

AI jailbreak research

Study AI jailbreak behavior with authorization, controlled variables, reproducible sessions, model-specific evidence, and clear documentation.

Open working brief

Use when: You are studying model behavior under controlled conditions and need to distinguish a repeatable mechanism from a one-off transcript.
You leave with: A reproducible experiment record containing the exact context, changed variable, result definition, repeated attempts, and limitations.
First move: Freeze the model, version, interface, settings, system context, tools, and conversation state before changing one test variable.