parametric-scribe
Cycleaddict/parametric-scribeEnables "Time Machine" coding. Records tasks as a Recipe and allows intelligent replay/modification of history.
SKILL.md
name: parametric-scribe description: Enables "Time Machine" coding. Records tasks as a Recipe and allows intelligent replay/modification of history.
Parametric Scribe
Overview
You are the Keeper of the Recipe. Unlike normal coding where "what's done is done," you maintain a Parametric History of the project in docs/recipe.yaml.
Mode 1: Recording (The default)
When: After completing any coding task.
Action: Append a new entry to docs/recipe.yaml.
- step_id: "<sequential-id>"
timestamp: "<iso-date>"
intent: "<short-summary>"
prompt: "<the-exact-prompt-used>"
output_files: ["<list-of-modified-files>"]
git_sha: "<current-head-sha>"
Mode 2: Time Travel (Replay)
When: The user says "Change Step X" or "Replay from Step X".
The Protocol:
- Read
docs/recipe.yaml. - Identify the Target Step (Step X).
- Hard Reset:
git checkout <Step X-1 SHA>(Go back to the state before Step X). - Execute Target: Run the New Prompt provided by the user.
- Intelligent Rebase (The Magic):
- For each Subsequent Step (Step X+1, X+2...):
- Read the Original Intent (Prompt).
- Look at the Current Codebase (which is now different).
- Self-Correction: "The original prompt asked to modify
app.py, but we are now inindex.js. I will apply the intent (Add Login) toindex.jsinstead." - Execute the adapted prompt.
- Update the
recipe.yamlwith the new file paths and SHAs.
- For each Subsequent Step (Step X+1, X+2...):
Mode 3: Forking
When: The user says "Create a variant."
Action: Copy recipe.yaml to recipe-variant-b.yaml and modify the target step there, leaving the original intact.
README
Parametric Scribe: Time Travel for AI Coding
Most AI coding is destructive. You paint over the canvas. If you realize your foundation was wrong, you have to scrape it off or start over.
Parametric Scribe changes this. It treats your coding session like a CAD Timeline or a Non-Linear Editor.
- Record: Every task you complete is saved as a discrete "Step" with an Intent (Prompt) and Outcome (Files).
- Replay: Want to change Step 1 from "Flask" to "Django"? The Scribe doesn't just change the text; it replays the subsequent steps, intelligently adapting them to the new foundation.
The Problem
In CAD, if you change a Sketch from a Square to a Circle, the Extrude feature updates automatically.
In Code, if you change app.py to index.js, your subsequent "Add Login" prompt fails because it's looking for Python code.
The Solution
Parametric Scribe uses the LLM as an Intelligent Compiler. During replay, it looks at the original intent of the subsequent steps ("Add Login") and applies it to the new reality (the Node.js app), effectively "healing" the broken references.
Usage
This is a Superpower Skill.
- Add this folder to your Agent's workspace.
- Tell the Agent: "Read
parametric-scribe/SKILL.md." - Start coding.
The Recipe File
Your history is saved in docs/recipe.yaml. This is your Source of Truth.