lisa-loops-memory

Lisa

Give Claude Code a memory before it touches your code.

Lisa is a Claude Code plugin that reads your project before acting on it — extracting tech stack, constraints, prior decisions, and open work into structured files that every future Claude session can instantly use. No more re-explaining what the project is. No more "wait, why was this built this way?" No more lost context.

---

The best way to try it

You know that project you started six months ago and never quite finished? The one with the half-written README, the TODOs in the code, the Slack thread where you worked out the architecture? Open it in Claude Code and run:

/lisa:rescue

Lisa will read your git history like an archaeologist, reconstruct what you were building and why you stopped, understand your tech stack and constraints, generate a fresh roadmap, and create structured work items ready to pick up. In one command, a dormant project becomes a living, organized one — with a memory file that any future Claude session can load in seconds.

---

What it actually does

Lisa runs a staged pipeline. Each stage produces files that the next stage (and future Claude sessions) can read:

Stage	Command	What you get
0 — Research	/lisa:research	Git archaeology report: original mission, timeline, why it stalled, rescue recommendation
1 — Discover	/lisa:discover	.gt/memory/semantic.json — tech stack, constraints, personas, non-goals
2 — Plan	/lisa:plan	scopecraft/ — vision, phased roadmap, epics + user stories, risks, metrics, open questions
3 — Structure	/lisa:structure	.gt/beads/ + .gt/convoys/ — individual work items and bundled assignments
5 — Reconcile	/lisa:reconcile	Cross-project alignment report — checks multiple repos stay consistent

Two composite shortcuts:

/lisa:migrate    # Stages 1–3: discover + plan + structure (for active projects)
/lisa:rescue     # Stages 0–3: full rescue pipeline (for abandoned projects)

---

Install

/plugin marketplace add auge2u/lisa-helps-ralph-loops
/plugin install lisa@lisa-helps-ralph-loops

That's it. No API keys, no database, no cloud. All outputs are local files.

---

What you get in your repo

After /lisa:migrate or /lisa:rescue, your project has:

your-project/
├── .gt/
│   ├── memory/
│   │   └── semantic.json      ← everything Claude needs to understand the project
│   ├── beads/
│   │   └── gt-abc12.json      ← individual work items with acceptance criteria
│   └── convoys/
│       └── convoy-001.json    ← beads grouped for assignment
└── scopecraft/
    ├── VISION_AND_STAGE_DEFINITION.md
    ├── ROADMAP.md
    ├── EPICS_AND_STORIES.md
    ├── RISKS_AND_DEPENDENCIES.md
    ├── METRICS_AND_PMF.md
    └── OPEN_QUESTIONS.md

The semantic.json is the key file. Point any future Claude session at it and you skip the 20 minutes of re-orientation. It's a structured summary of what the project is, what it's built with, what constraints exist, and what's been decided — all in a format Claude reads natively.

---

Quality gates

Every stage validates its own output against 31 declarative gates before you move on. If something is missing or wrong, you know before the next agent touches it.

python3 plugins/lisa/hooks/validate.py --stage discover
python3 plugins/lisa/hooks/validate.py --stage all

No PyYAML? It runs in fallback mode (file and JSON checks). Install PyYAML for full pattern validation.

---

Part of a larger ecosystem

Lisa is the foundation layer of a three-plugin ecosystem:

Plugin	What it adds	Status
Lisa (this)	Pipeline, memory, work structure	Alpha v0.3.0
Carlos	Specialist analysis: market fit, tech debt, roadmap depth	Beta v1.2.0
Conductor	Multi-agent orchestration and task tracking	GA v1.0.0

Each plugin works completely standalone. Install one, or all three — the ecosystem is additive. Lisa's outputs (.gt/ files and scopecraft/) are the shared language between them.

---

Why this exists

Claude Code is powerful, but every new session starts cold. You re-explain the project, re-establish context, re-decide things that were already decided. Lisa solves the cold-start problem by giving Claude a structured memory of the project that persists across sessions.

It's especially useful for:

• Abandoned projects — reconstruct context and get a clear path to resume
• Onboarding — new contributors (human or AI) get up to speed immediately
• Multi-agent workflows — structured work items that agents can pick up without ambiguity
• Long-running projects — semantic memory keeps Claude oriented as the project grows

---

License

MIT