AI protocol · ChatGPT Projects · append-only continuity

Keep the shape.
Lose the drift.

Memory-Assisted Shaping keeps long idea-shaping sessions coherent without turning notes into the work surface.

View repository Use with ChatGPT Projects
01 OP input

One idea enters the session.

02 GPT shaping

Gates, decisions, and boundaries stay visible.

03 Minimal signal

Only continuity-critical material is retained.

04 Python append

Persistence is append-only and model-independent.

05 Clean artifact

Produced only after explicit approval.

The failure mode

Long chats do not fail all at once. They drift.

Examples harden into decisions. Discarded paths return. Open gates vanish. Final artifacts start carrying process residue.

Without Memory-Assisted Shaping

Conversation residue leaks into the artifact.

  • Examples become structure.
  • Discarded options return later.
  • Open questions look resolved.
  • Final output carries shaping noise.

With Memory-Assisted Shaping

The session keeps operational continuity.

  • Decisions stay recoverable.
  • Real gates remain explicit.
  • Source boundaries are protected.
  • Final artifacts stay clean.

Core mechanic

GPT shapes. Python persists.

GPT emits small memory signals only when continuity would degrade without retention. Python timestamps and appends. The model does not own, normalize, or rewrite memory files.

Real gates

Only points that change scope, source of truth, runtime, lifecycle, or artifact correctness become gates.

Discarded paths

Rejected directions stay recoverable so they do not silently re-enter the session.

Source boundaries

Declared material, inference, and prior memory do not collapse into the same authority layer.

Artifact separation

The final artifact is clean, standalone, and produced only after explicit approval.

✅ Use it when

  • one idea needs more than a few turns
  • you are shaping a protocol, workflow, spec, or handoff
  • decisions and open gates matter
  • you need recap or re-entry without drift

🚫 Do not use it when

  • you only need a quick answer
  • you want a full note-taking system
  • you want to store everything
  • you want memory to become authority
Package

Three files. One boundary.

protocol.md

Behavioral contract for shaping one active idea.

note-io.md

Boundary between GPT signals and Python persistence.

notes.py

Small stdlib-only helper for append-only session notes.

Use with ChatGPT Projects

Add the files. Tell GPT the mode. Shape normally.

Ask for a recap when state matters. Ask for a final artifact only when the shape is ready.

Open repository
Use Memory-Assisted Shaping for this idea.