Open source
Build your next research workflow.
Adapt a Python example, explore its data, and share your work. Use your coding agent or run it yourself.
Free and open source. Offline examples need no account or API key.
Public preview
Agent Lab is open on GitHub. Bring an example, a visualization, or an improvement to an existing workflow. Each contribution should include the inputs and instructions someone else needs to run it.
Watchlist Desk is a local Docker prototype. It shows changes between fictional snapshots and flags missing or incompatible data.
Run the Docker example GitHub. Opens in a new tab.Experiments
Three Python workflows with sample data. Run them offline without an account, API key, model, or extra Python packages.
Watchlist Investigator
Read GEX, VEX and flow observations together, with source times and missing data kept visible.
Fictional defaults. Requires a separate, pinned Agent Kit checkout.
Workflow guide GitHub. Opens in a new tab.Journal Reviewer
Review closed paper-trade rows and the notes you entered. Start with the sample, then bring a supported CSV.
Closed USD cash-equity paper trades. No inferred strategy or tax analysis.
Workflow guide GitHub. Opens in a new tab.Market Brief
Turn dated entries from one press-release feed into a compact brief. Start offline, then use saved RSS or an explicit public-feed fetch.
Selects 1–20 entries from a single feed. Not a market-wide scan.
Workflow guide GitHub. Opens in a new tab.Your first run
With Git and Python 3.11+, clone the Lab and run Market Brief. This first run uses fictional data, makes no network requests and spends no API credits.
Read the getting-started guide GitHub. Opens in a new tab.
git clone https://github.com/SkylitAI/skylit-agent-lab.git
cd skylit-agent-lab
python3 -X utf8 -I -B experiments/market-brief/run.pyThe command prints paths to a Markdown report and a run record. Open both to inspect the result and its inputs. On Windows, use py -3 if python3 is unavailable.
Build with others
Contribute a workflow, data adapter, visualization, documentation improvement or reproduction. Make it possible for someone else to run your work and understand its limits.
Contribution guide GitHub. Opens in a new tab.Give the experiment a clear question, owner, inputs and expected output.
Include a reproducible example, sources, known gaps and resource costs. Keep credentials and private reports local.
Submit the experiment for review. Useful, reusable work can graduate into Agent Kit through a reviewed contribution.
Lab is the home for experiments and community contributions. Agent Kits provides maintained helpers, starter workflows and setup guides. Lab experiments can use Kit components and later contribute improvements back.
The Lab has guidance for Codex, Claude Code, OpenClaw, other skill-capable hosts, MCP-only hosts and standalone Python, with provisional Muse Code guidance. Python examples are tested; host-driven runs remain unverified. See the agent guide GitHub. Opens in a new tab.
Yes. Agent Lab and Agent Kit are public under the MIT license. You can clone them, make changes and submit a pull request. Skylit services and agent subscriptions have separate terms and costs.
No. These seeds produce research reports from explicit inputs. They do not submit orders, manage a brokerage account or promise investment returns. Live-service and agent-host certification remain pending.
Contribute
For research and education. These workflows do not place trades. Live services and agent hosts have not been independently verified.