Splice

Share orchestrated context between Kiro, Cursor, and Claude Code — one layer above the assistant.

Splice is a thin meta-tooling layer for any codebase. It ingests errors, diffs, and files once, routes to the right agent profile (debugger, architect, implementer), and hands off a reduced context bundle between IDEs via a shared session in your repo.

Three orchestration modes: mock (offline) · Ollama (local, privacy-first) · Bedrock (cloud).
No AWS required for IDE hooks.

Install

pip install splice-cli

Then set up your project:

cd your-project
splice setup        # detects installed IDEs and copies hooks automatically
splice doctor       # verify config

Restart Kiro, Cursor, and/or Claude Code, then:

/harness debug the failing test in src/auth.ts

Why this exists

If you use Kiro, Cursor, and/or Claude Code on the same project, you probably:

• Paste the same error log twice
• Lose session continuity when switching IDEs
• Burn tokens on README noise and unrelated files
• Worry about sending proprietary code to cloud orchestrators

Splice fixes that with a single trigger: /harness

Quick start (without pip)

git clone https://github.com/hsaenzG/splice.git /tmp/splice
bash /tmp/splice/scripts/install-splice.sh /path/to/your-repo

Privacy-first: Ollama (recommended for local work)

Your orchestration code and context never leave your machine:

brew install ollama          # or https://ollama.com
brew services start ollama
ollama pull llama3.2

Edit .splice/config.json:

{ "orchestrator": "ollama", "ollama": { "model": "llama3.2" } }

splice doctor
/harness debug my issue        # in Kiro, Cursor, or Claude Code

Kiro/Cursor/Claude Code may still use their own cloud models when you chat. Splice controls only the orchestration step.

Daily workflow

# 1. Capture test/build output (optional)
npm test 2>&1 | splice capture-error

# 2. Triage in Kiro or Claude Code
/harness debug the auth failure

# 3. Implement in Cursor or Claude Code (same repo — shared .splice/session.json)
/harness apply the fix using the orchestrated bundle

# 4. Inspect bundle + metrics
splice status
cat .splice/active-bundle.md

How it works

flowchart LR
    subgraph IDEs
        K[Kiro]
        C[Cursor]
        CC[Claude Code]
    end

    subgraph SpliceLayer["Splice"]
        I[Ingest] --> O{Orchestrator}
        O -->|mock| B[Bundle]
        O -->|ollama| B
        O -->|bedrock| B
    end

    S[(.splice/)]

    K -->|/harness| I
    C -->|/harness| I
    CC -->|/harness| I
    O --> S
    B --> K
    B --> C
    B --> CC

CLI reference

splice setup                                   # install IDE hooks into current project
splice setup /path/to/project --ide kiro       # specific project / specific IDE
splice init --assistant kiro|cursor|claude     # create or resume a session
splice orchestrate --assistant kiro --prompt "/harness ..."
splice handoff --from kiro --to cursor
splice capture-error                           # pipe stderr/stdout → .splice/last-error.log
splice doctor                                  # verify mock/ollama/bedrock
splice status                                  # show active session

Choose your orchestrator

Env override: SPLICE_ORCHESTRATOR=ollama

If Ollama is down, Splice falls back to mock automatically.

Configure your project

.splice/config.json (created automatically by splice setup):

{
  "project": "my-app",
  "orchestrator": "ollama",
  "captureGitDiff": true,
  "includePaths": ["src/**/*.ts"],
  "errorSources": [".splice/last-error.log"],
  "ollama": { "model": "llama3.2", "baseUrl": "http://localhost:11434" },
  "bedrock": { "modelId": "amazon.nova-lite-v1:0" },
  "promptPathPatterns": true
}

AWS deploy (optional)

chmod +x scripts/deploy.sh && ./scripts/deploy.sh

Stack sets SPLICE_ORCHESTRATOR=bedrock on Lambda. Local hooks don't need AWS.

Project structure

├── splice_cli/              # pip-installable package
│   ├── cli.py               # splice entrypoint
│   ├── shared/              # orchestrator, collect, store, config
│   └── data/                # hooks + templates bundled with the package
├── scripts/
│   ├── install-splice.sh    # legacy bootstrap (pre-pip)
│   ├── splice-cli.py        # legacy CLI script
│   └── demo-dual-ide.py
├── integrations/hooks/      # Kiro + Cursor + Claude Code hook scripts
├── src/shared/              # same modules for AWS Lambda handlers
├── ORCHESTRATORS.md
├── INSTALL.md
└── demo-app/                # Optional workshop sample

Requirements

• Python 3.10+
• Kiro, Cursor, and/or Claude Code
• Ollama (optional, for local LLM orchestration)
• Git (optional, for diff capture)

Docs

Contributing

PRs welcome — especially IDE adapters (Windsurf, Zed) and Ollama model presets.

License

MIT — see LICENSE.

Author

Created by Hazel Saenz — LATAM Developer Advocate at AWS | Former AWS Serverless Hero.