inner-loop-os

A myth-bound bootloader for LLMs
CLI thinking for LLMs: file-based truth, deterministic flow, portable across models.
Drop the repo into context, type INIT, and ship.

How It Works

Drop this repo (or paste its files) into any model's context, type INIT, and the agent will:

Load identity Dr. X identity contract (tone, cognition, modes)
Inspect state Filesystem state (GOAL.md, STATUS.md, PRD.md, UX.md)
Determine step Next concrete action with maximum momentum
Execute Minimal questions, maximum execution
The One Ritual
  1. Provide context (repo or relevant files)
  2. Type:
INIT

The system either resumes from state or asks 4 questions max.

Modes (Semantic Switches)

Set a mode explicitly when needed:

MODE: DEBUGGER
INIT
ARCHITECT Structure, interfaces, constraints (default)
CREATOR Naming, synthesis, momentum
DEBUGGER Isolate issues, precision questions
WRITER Docs that execute, crisp articulation
GHOST Minimal output, artifacts only

Recommended Workflows

Start a new challenge
  1. Copy templates into /work
  2. Fill work/GOAL.md
  3. Create work/STATUS.md with state = DEFINING
  4. Run INIT
Resume work (fast)
  1. Update work/STATUS.md (blocker + next step)
  2. Run INIT
High-stakes choice
  1. Complete system/DECISION.md
  2. Apply decision
  3. Update work/STATUS.md
  4. Run INIT
After shipping or failing
  1. Complete system/RETRO.md
  2. Apply the patch
  3. Update work/STATUS.md
  4. Run INIT

Deterministic Boot Process

On INIT, the system searches for control files using this path priority:

1. ./work/
2. ./ (repo root)
3. ./system/
4. ./templates/

Then inspects files in this order:

System Resources

Create work/GOAL.md and work/STATUS.md, then type INIT.