Skip to content
 
 

Latest commit

 

History

84 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Lisa Loop

Without structure, AI agents vibe-code. They write code and tests from the same understanding, so bugs and tests share the same blind spots. There's no audit trail, no review gate, no way to tell if the answer is trustworthy or just plausible.

Lisa is a CLI tool that orchestrates Claude Code through a structured engineering process — scoping before coding, independent verification, iterative refinement, and human review at every stage.

How It Works

  lisa init → scaffold .lisa/ + ASSIGNMENT.md
                    │
                    ▼
  ┌───────────────────────────────────────────────┐
  │  SCOPE (Pass 0)                               │
  │  Define methods, acceptance criteria, staged   │
  │  plan — no code yet                           │◄── human refine loop
  └──────────────────┬────────────────────────────┘
                     ▼
  ┌───────────────────────────────────────────────┐
  │  INDEPENDENT VERIFICATION (one-time prologue) │
  │  Test criteria from literature, before code    │
  └──────────────────┬────────────────────────────┘
                     ▼
  ┌───────────────────────────────────────────────┐
  │  SPIRAL PASSES (Pass 1..N)                    │
  │                                               │
  │  ┌─────────┐  ┌─────────┐  ┌────────┐        │
  │  │ REFINE  │→ │  BUILD  │→ │VALIDATE│        │
  │  └─────────┘  └─────────┘  └────────┘        │
  │                                               │
  │  Each pass widens scope & tightens tolerances │◄── human review gate
  └──────────────────┬────────────────────────────┘
                     ▼
  lisa finalize → answer.md + report.md

Scope locks down the methodology, acceptance criteria, and a staged plan — before any code is written. A human reviews and refines until satisfied.

Independent Verification runs once after scoping. A separate agent reads reference literature and writes verification scenarios (markdown, not code) that define what "correct" looks like — independently from the agent that will write the implementation.

Spiral Passes iterate through Refine → Build → Validate. Each pass increases fidelity: early passes get the structure right, later passes tighten tolerances and handle edge cases. After every pass, the human decides: accept, continue to the next pass, or redirect with guidance.

Grounded in Engineering Practice

Lisa combines two established frameworks, adapted for AI-assisted work.

The V-Model says: define what "correct" means before you write code, then validate against those pre-defined criteria. In Lisa, the left arm of the V is the scoping phase (acceptance criteria) and the independent verification prologue (test scenarios from literature). The right arm is the Validate phase, which executes those pre-defined scenarios against the implementation. The criteria exist before the code does — so they can't be shaped by implementation assumptions.

The Design Spiral says: iterate with increasing fidelity, with human control at each turn. Lisa's spiral passes do exactly this — each pass refines the approach, builds with more detail, and validates against tighter standards. The human reviews every pass and controls when to stop.

Independent Verification

When the same agent writes both code and tests, errors correlate — the tests pass because they share the code's assumptions, not because the code is correct.

Lisa breaks this correlation by separating three roles:

  DDV Agent (opus)          Build (sonnet)          Validate (opus)
  ┌───────────────┐        ┌───────────────┐       ┌───────────────┐
  │ Read papers    │        │ Read scenarios │       │ Read scenarios │
  │ Write scenarios│───────→│ Write code     │──────→│ Write tests    │
  │ (no code)      │        │ (no DDV tests) │       │ Run all tests  │
  └───────────────┘        └───────────────┘       └───────────────┘
    ▲ independent              ▲ separated              ▲ verified

The DDV Agent reads authoritative sources and writes verification scenarios — expected values, edge cases, domain constraints — as markdown. The Builder writes implementation code but never touches the verification tests. The Validator converts scenarios into executable tests and runs the full suite.

Disagreements between the DDV scenarios and the implementation are signals, not bugs to suppress. They surface misunderstandings that would otherwise hide until production.

This works for any domain with authoritative sources and testable expected values: physics, econometrics, regulatory standards, engineering benchmarks.

Cornerstones

  1. Traceable methodology. Every methodological choice traces to an authoritative source. No equation without a paper. This makes results defensible and reviewable — not just plausible.

  2. Auditable engineering judgment. Assumptions, sanity checks, and parameter choices are written down, versioned, and executed — not left as implicit decisions buried in code comments.

  3. Visual evidence for review. Plots, comparison charts, and diagrams surface problems that tables and numbers hide. Visual artifacts are the preferred way to present results for human review.

  4. The history is the deliverable. The refinement record across spiral passes — not just the final answer — is evidence of trustworthiness. Every pass is preserved as a complete record of how the solution evolved.

Getting Started

Prerequisites

  • Claude Code installed and authenticated
  • Git configured with user.name and user.email

Install

Download the latest release binary (Linux x86_64):

curl -fsSL https://github.com/freol35241/lisa-loop/releases/latest/download/lisa-linux-x86_64 -o lisa && chmod +x lisa && sudo mv lisa /usr/local/bin/

Or build from source:

cargo install --path .

Usage

lisa doctor                  # Check prerequisites
lisa init                    # Scaffold .lisa/ and ASSIGNMENT.md
# Edit ASSIGNMENT.md with your problem description
# Add reference papers to .lisa/references/core/
lisa run                     # Run the full spiral

Commands

lisa run                     # Full spiral: scope → DDV → passes → finalize
lisa run --max-passes 3      # Limit spiral passes
lisa scope                   # Pass 0 only (scoping)
lisa resume                  # Resume from saved state
lisa status                  # Print current spiral state
lisa history                 # Pass-by-pass history
lisa rollback <pass>         # Roll back to a pass boundary
lisa continue "<question>"   # Follow-up question after acceptance
lisa finalize                # Produce answer.md + report.md
lisa eject-prompts           # Copy prompts to .lisa/prompts/ for customization
lisa doctor                  # Check environment

Configuration lives in .lisa/lisa.toml (models, limits, review gates, paths, commands). Run lisa init to see the full default config with comments.

Human Interaction

Pass Review Gate

═══════════════════════════════════════════════════════
  SPIRAL PASS N COMPLETE — REVIEW REQUIRED
═══════════════════════════════════════════════════════

  Answer:      142.3 kN total resistance
  Progress:    Δ 12% from prev
  Tests:       DDV: 8/8 | Software: 15/15 | Integration: 2/2
  Agent recommends: CONTINUE

  [A] ACCEPT — produce final report
  [C] CONTINUE — next spiral pass
  [R] REDIRECT — provide guidance

Scope Review Gate

After Pass 0, review methodology and acceptance criteria before any code is written. Options: Approve, Refine (agent re-runs with feedback), Edit (modify files directly), or Quit.

Credits

Lisa Loop extends the Ralph Wiggum technique by Geoffrey Huntley — an approach to iterative AI coding where an outer agent loop drives Claude Code through repeated build-test cycles. Lisa adds scoping, independent verification, human review gates, and structured methodology on top of that foundation. Named after Lisa Simpson, the rigorous counterpart to Ralph Wiggum.

About

A methodology and toolbox for solving complex engineering and research problems with AI agents, grounded in peer-reviewed literature, with explicit verification, validation, and convergence tracking.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages