# SensecraftXStudio

Name: SensecraftXStudio
Type: Agent behavior contract
URL: https://xxyoudeadpunkxx.github.io/SensecraftXStudio/
Repository: https://github.com/XxYouDeaDPunKxX/SensecraftXStudio
Runtime: AI-assisted technical workspaces using AGENTS.md
License: CC BY-SA 4.0
License URL: https://creativecommons.org/licenses/by-sa/4.0/
Canonical file: AGENTS.md

## What this is

SensecraftXStudio is an agent behavior contract for AI-assisted technical work.

It gives the assistant a compact working frame before it acts: close the target, read surrounding context, surface expansion, preserve authority, and report what is actually grounded.

It is not a guarantee of correctness, not a replacement for technical judgment, and not a general personality prompt.

## Operating line

Confidence is not trusted here. Close the frame before touching the work.

## Use case

Use SensecraftXStudio when an assistant needs to inspect, edit, review, or reason about real project material and must avoid plausible drift, premature edits, unsurfaced expansion, or weak closure.

## Canonical source

The canonical file is:
AGENTS.md

Treat the repository as the live public source. Treat this file as the LLM-readable project index, not as the full contract specification.

## Purpose

Keep the assistant cognitively bounded before it acts.

The contract pushes the session to close the real object, inspect the surrounding system, surface expansion, and close with a grounded task-state readout.

## Core problem

AI can stay plausible after the frame has already drifted.

## Use cases

- repository or workspace analysis
- implementation inside an existing codebase
- technical review
- structured comparison
- decision support
- verification before closure

## Activation

Copy AGENTS.md into the target project or workspace.

Start a fresh assistant session in an environment that can read workspace files.

Ask the assistant to read AGENTS.md in full before acting.

Tell it to use AGENTS.md as the operative working frame for the task.

Then give the real task.

Minimal operator instruction:
Read AGENTS.md in full and use it as the operative frame before acting on this workspace.

## Visible before action

- the obvious move
- the real object being touched
- the surrounding context that gives the move meaning
- the valid alternative readings
- whether the move stays contained or expands scope, risk, or structure
- why the selected move is the smallest correct one

## Final response contract

When a response includes a change, recommendation, or finding with operational consequence, it should close with:

- Touch: what changed and what stayed untouched within expected scope
- Ground: what the move or conclusion was grounded on
- State: what is verified, inferred, unresolved, or not inspected
- Convergence: whether the task is converged, divergent, or blocked

## Stop conditions

- missing or conflicting context
- real object not closed
- authority not closed
- drift from the operative contract
- unsurfaced scope, risk, or structure expansion
- destination ambiguity that changes meaning or consequence
- multiple valid paths without justified closure
- incoherent current state that cannot support forward fixing

## Limits

SensecraftXStudio does not make AI deterministic.

It does not replace technical judgment.

It does not guarantee that a host will preserve context, priority, or workspace access perfectly.

## Discovery Set

Machine-readable project files:

- llms.txt: LLM-readable project index.
- raw-manifest.json: structured machine-readable project manifest.
- robots.txt: crawler access declaration.
- sitemap.xml: sitemap for the GitHub Pages surface.

HTML discovery signals:

- canonical URL
- rel=alternate for llms.txt
- rel=alternate for raw-manifest.json
- JSON-LD SoftwareSourceCode metadata
- Open Graph metadata
- footer machine-readable links

## Keywords

AI-assisted technical work, AGENTS.md, working frame, operational contract, technical review, workspace analysis, bounded AI behavior, grounded closure, SensecraftXStudio, agent behavior contract, plausible drift, task-state closure