by eddyzzl
MARVIS-Agent: all-purpose credit risk agent for model development, validation, data processing, feature engineering, and strategy workflows.
# Add to your Claude Code skills
git clone https://github.com/eddyzzl/marvis-risk-agentGuides for using ai agents skills like marvis-risk-agent.
marvis-risk-agent is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by eddyzzl. MARVIS-Agent: all-purpose credit risk agent for model development, validation, data processing, feature engineering, and strategy workflows. It has 53 GitHub stars.
marvis-risk-agent's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/eddyzzl/marvis-risk-agent" and add it to your Claude Code skills directory (see the Installation section above).
marvis-risk-agent is primarily written in Python. It is open-source under eddyzzl on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh marvis-risk-agent against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
MARVIS-Agent is built for governed credit-risk work that should stay close to local files, local runtimes, and auditable evidence. It covers model development, model validation, data processing, feature derivation, feature analysis, strategy generation, strategy validation, monitoring, and governed task automation.
The current V1.1.6 release ships model validation as the first stable built-in workflow. It can run notebook-based validation tasks, generate structured evidence, and draft Excel/Word validation reports through Agent mode. Model validation is the first workflow, not the product boundary.
The current V1.1.x line already ships the Agent Memory Foundation for historical validation metric comparison. Planned next steps are documented in docs/roadmap.md: V2 adds the Agent Plugin/Tool Runtime, and later releases build modeling and strategy capability packs on that runtime.
MARVIS-Agentmarvis/static/brand/Private or customer-specific branding is intentionally not committed. To apply a local brand, create an ignored workspace config:
workspace/branding/brand.json
Example:
{
"platform_name": "本地信贷风控智能体",
"browser_title": "本地信贷风控工作台",
"primary_color": "#1f6feb",
"logo": "private-logo.svg",
"favicon": "private-logo.svg"
}
Put referenced logo files next to brand.json. When workspace/branding/ is absent, the app falls back to the public MARVIS brand.
See docs/branding.md for details.
pypmml if you need PMML scoring.Clone the repository, then install from the checkout. Create an environment with any name you prefer. For example, with venv:
git clone https://github.com/eddyzzl/marvis-risk-agent.git
cd marvis-risk-agent
python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -e ".[dev]"
Or with conda:
git clone https://github.com/eddyzzl/marvis-risk-agent.git
cd marvis-risk-agent
conda create -n marvis python=3.12
conda activate marvis
python -m pip install -U pip
python -m pip install -e ".[dev]"
After installation, start MARVIS with:
marvis
By default, this is equivalent to:
marvis serve --host 127.0.0.1 --port 8000 --workspace ./workspace
Then open http://127.0.0.1:8000/.
The Python module name marvis is retained in V1 for compatibility with the current validation runtime. The older entrypoints still work:
python -m marvis serve --host 127.0.0.1 --port 8000 --workspace ./workspace
marvis-risk-agent serve --host 127.0.0.1 --port 8000 --workspace ./workspace
When creating a task, the material directory must be under the current workspace or the current user's home directory by default. On Windows, allow another drive or local folder before startup:
$env:RMC_MATERIAL_ROOTS="D:\model_materials"
marvis serve --host 127.0.0.1 --port 8000 --workspace .\workspace
When running under WSL2, enter the WSL path such as /mnt/c/Users/<you>/Downloads/project, not a C:\... Windows path.
When running multiple worktrees at the same time, use different ports and different workspaces. Profiles choose safe defaults:
# Stable main demo
marvis serve --profile main
# http://127.0.0.1:8000, workspace ./workspace-main
# V1.1 development or comparison
marvis serve --profile v1-1
# http://127.0.0.1:8001, workspace ./workspace-v1-1
Explicit options override profile defaults:
marvis serve --profile v1-1 --port 8017 --workspace ./custom-workspace
If MARVIS was installed from a GitHub clone and the checkout is on a clean main branch, run:
marvis update
The command runs git fetch origin, git pull --ff-only origin main, then refreshes the editable install:
python -m pip install -e .
If tracked local files have uncommitted changes, marvis update refuses to continue. Commit, stash, or back up those tracked changes before updating. Untracked local files are allowed unless Git itself detects that a pull would overwrite them.
If your current older install does not have marvis update yet, run one manual upgrade from the repository directory:
git pull --ff-only origin main
python -m pip install -e .
After that, future upgrades can use marvis update.
python -m pytest -q
ruff check marvis tests --extend-exclude '*.ipynb'
node --check marvis/static/app.js
Use the release helper instead of raw git push when publishing a new public version. Run it after the feature, fix, or documentation changes have been verified and committed. The helper requires a clean worktree and creates a separate version bump commit plus an annotated tag.
python scripts/release_push.py --bump patch
The helper updates release metadata, creates a release commit, creates an annotated Vx.y.z tag, and pushes main plus the tag. See docs/versioning.md for the full release sequence and versioning rules.
This project is released under the MIT License. See LICENSE for details.