Maintainer of this skill?
Automated security scan — not a certification.
A Windows workbench and native Python Pi harness for Odoo 19: approved business actions, state readback, traceable runs, and ERP-Bench evaluation.
# Add to your Claude Code skills
git clone https://github.com/Sanssssssssssssssss/ERP-agentLast scanned: 10/20/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-10-20T09:10:09.314Z",
"npmAuditRan": true,
"pipAuditRan": true,
"promptInjectionRan": true
}See how ERP-agent compares with popular alternatives.
ERP-agent is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by Sanssssssssssssssss. A Windows workbench and native Python Pi harness for Odoo 19: approved business actions, state readback, traceable runs, and ERP-Bench evaluation. It has 157 GitHub stars.
Yes. ERP-agent passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/Sanssssssssssssssss/ERP-agent" and add it to your Claude Code skills directory (see the Installation section above).
ERP-agent is primarily written in Python. It is open-source under Sanssssssssssssssss 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 ERP-agent against similar tools.
No comments yet. Be the first to share your thoughts!
30 days in the Featured rail · terms & refunds
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
Automated security scan — not a certification.
“Confirm S00012 if its untaxed amount is under 5,000 CNY. Leave S00009 alone. No shipment or invoice.”
The agent reads the order, requests approval for the exact change, executes the approved action, and reads the result back. The execution desk keeps the goal, current action, approvals and evidence together; chat stays alongside it.
The condensed demo shows recorded states from a real model/API run against an isolated Odoo test company. Validation results distinguish this desktop workflow from the three ERP-Bench tasks used to check the integrated backend.
Sales · Purchasing · Inventory · Manufacturing · Invoicing · Payments · Refunds · Reconciliation
Business state, approvals and action receipts persist across interruptions. An uncertain write requires reconciliation before further execution. Tool availability follows the current task; field access and action approval remain explicit.
Bring your own Odoo 19 JSON-2 connection and model endpoint.
The redesigned desktop shown above is available from source. The published Windows v0.6.0 preview contains the earlier UI; release notes describe that build.
Requirements: Node.js 22+, Python 3.13 and uv.
git clone https://github.com/Sanssssssssssssssss/ERP-agent.git
cd ERP-agent
uv sync --locked
cd desktop
npm ci
$env:WORKBENCH_HOST_ROOT = (Resolve-Path ..).Path
$env:WORKBENCH_PYTHON = (Resolve-Path ..\.venv\Scripts\python.exe).Path
npm run dev
Open Connection settings, enter Odoo and model credentials, and start a new chat. Attach CSV, TXT, XLSX, PDF or images as reference material. Settings include dark appearance, reduced motion and chat storage location. Printed-document extraction runs locally; spreadsheet formulas are not evaluated.
For a reproducible local company, see the persistent demo and 10,000-document validation fixture. The sample orders.csv specifically matches the ERP-Bench 2262 seed.
npm run typecheck
npm run self-check
npx playwright install chromium
node scripts/renderer-check.mjs
npm run check:execution
npm run check:trace
Backend checks and coverage are indexed in TESTING.md. Packaging instructions, pinned sidecar inputs and rollback commands are in the development guide.
The desktop runs the native Python Harness with dynamic tools, controlled business procedures and recorded world state. It connects directly to Odoo; historical MCP comparison code remains in the research tree. Identity-scoped BM25 knowledge search is integrated. Long-term memory is optional and defaults to off; the separate vector RAG experiment is not connected to the desktop.
The workflow families above have local test coverage; specific acceptance and remaining limitations are documented in the enterprise ledger, manual business results and backend validation. These fixtures do not establish production bank, tax or HR integration. A posted invoice is not proof of payment, and a verified Odoo state is not a benchmark score.
ERP-Bench and Harbor provide a separate evaluation path with pinned tasks, snapshots and scoring. Mock and renderer checks verify UI mechanisms; real API runs and Odoo readbacks verify the recorded test workflows.
Tool recovery validation records the failure causes, local API decisions, fresh business runs and token usage for each case.
| Path | Purpose |
|---|---|
| desktop/ | Electron application, React UI, IPC and packaging |
| src/erp_harness/app/ | Sessions, execution, approvals, materials and business views |
| src/erp_harness/erp/ | Native Odoo reads, writes, policy and verification |
| src/erp_harness/ | Agent runtime, providers, context and tools |
| bench/ | ERP-Bench, Harbor adapters and frozen configurations |
| experiments/ | Reproducible controls, evidence and research notes |
Original integration code is MIT. Adapted components retain their own licenses and pinned provenance in sources.lock.json, THIRD_PARTY_NOTICES.md and UI source notices.
See CONTRIBUTING.md, SECURITY.md and the historical lab entry.