Advanced Prompt Engineering

An execution-backed laboratory for designing, testing, falsifying, and publishing advanced prompting methods.

➡️ Start with README_START_HERE.md.

Repository contract

one repository = one authoritative prompt-research corpus
branches       = proposal and experiment spaces
main           = accepted working state
tags           = exact prompt and benchmark anchors
raw inputs     = inert, untrusted evidence
receipts       = reconstructable records of work performed
audits         = append-only observations bound to exact targets
AI agents      = replaceable, attributed research seats
human          = sole canonical promotion authority

This repository separates prompt text, hypotheses, experiments, evidence, evaluation, disposition, and promotion. A persuasive report does not promote itself. A passing demonstration does not automatically establish generality, novelty, safety, or production readiness.

First research object

prompts/adversarial-software-corpus-audit/v0.1.0/ contains the initial candidate prompt and its manifest. experiments/EXP-0001/ defines the first controlled evaluation programme around it.

Core commands

python3 scripts/verify_repo.py

The verifier checks the agent entry files, status object, task routing, prompt manifests, and prompt-content hashes without third-party dependencies.