Systems Engineering Skills is an open source requirements management toolkit from Arc. Also called Arc Skills, it gives an AI assistant 100 defined engineering workflows and a local Model Context Protocol (MCP) server for reading files, mapping requirements, checking recorded relationships and exporting reviewable proposals. The code and skills are MIT licensed. You do not need an Arc account.
It is useful when you have a customer specification, requirements spreadsheet or verification evidence to work through with an assistant. The toolkit supports individual engineering tasks. Your team still owns the authoritative requirements record, baseline decisions and formal approval process.
What does the open source toolkit include?
The catalogue contains 22 requirements skills, with the remaining tasks covering architecture and budgets, interfaces, verification, assurance, safety, reliability, configuration and technical reviews. Each skill specifies inputs, engineering judgement, a deliverable and an acceptance check.
Scroll across the table to see each workflow and its output.
| Your input | Workflow | Reviewable output |
|---|---|---|
| Requirements spreadsheet | Clean and map requirements | Column mapping, retained source statements, counts, duplicate IDs and unresolved links |
| Customer specification | Extract customer requirements | Candidate obligations with source locators and open questions |
| Parent and child requirements | Review derivation and coverage | Gaps, unsupported additions and proposed corrections |
| Requirement and applicable ECSS edition | Review against inspected criteria | Findings with edition and locator, or an explicitly advisory review |
| Requirements and verification information | Build a verification matrix | Method mapping, missing criteria and coverage questions |
| Two baselines and recorded links | Compare changes and assess impact | Exact record differences and candidate effects requiring engineering judgement |
These are instructions for an assistant to follow. A skill for an FMEA or a design review does not mean that the Python server independently performs that engineering analysis. The assistant supplies the reasoning, while the local tools perform defined file and model operations.
How do the skills and local tools work together?
MCP connects the assistant to the toolkit. The server implementation exposes skill search and retrieval, reference reading, file inspection, bounded reads, requirements import proposals, model validation, model loading, model comparison, relationship tracing and new-file export.
The distinction matters for change impact. The comparison tool identifies records added, removed or modified by stable ID. The trace tool follows recorded relationships within explicit limits. An assistant can interpret those results, but neither operation proves that every affected subsystem or test has been found. Missing links remain missing evidence.
The local exchange model represents requirements, systems, tests, test plans, risks and properties. Typed links can record derivation, allocation and verification scope. Validation checks structure, endpoints, identifiers and cycles. It does not prove technical correctness, complete traceability or successful verification.
Which files can it read and export?
The documented readers support UTF-8 CSV and TSV, XLSX, uncompressed ReqIF, JSON, text, Markdown, searchable PDF and DOCX body text and tables. Exports create new JSON, CSV, XLSX, Markdown or text files and refuse to overwrite existing files.
Scanned PDFs need a separate OCR step. PDF figures and image tables need separate inspection. Legacy XLS, macro-enabled XLSM and compressed ReqIF are not built in. Reading ReqIF does not establish lossless round-trip exchange with another requirements tool, and there is no ReqIF exporter.
Imports preserve source identifiers, original wording and provenance, including source file hashes. Mapping remains explicit, with rejected records and unresolved links visible. A generated proposal is a review artifact, not an approved import into a production system.
How do you get started?
Download the Systems Engineering Skills repository, then follow its setup guide for your client. The documented local route requires Python 3.11 or later and uv. It installs locked dependencies and connects a local MCP server to a dedicated work folder. Keep source material and working outputs separate from the toolkit installation.
The repository documents local setup for Codex and other stdio MCP clients, including Claude Desktop. It also describes conversation-scoped use of the attached skill instructions in ChatGPT where ZIP and file analysis are available. That attachment route does not install the local MCP server. Available file tools and setup steps depend on the host, so use the current guide and confirm actual tool discovery.
Start with the bundled synthetic CSV before using project data. A useful first request is:
Use Arc Skills to clean and map customer-requirements.csv. Preserve every source ID and original statement. Show the field mapping, duplicate IDs, missing information and unresolved links. Reconcile all row counts and export the proposal to a new file.
Check the result against the input yourself. Confirm the row count, inspect how parent relationships were interpreted and make sure unresolved values remain visible. A successful connection proves access to tools. A reviewed example establishes whether the workflow fits your task.
What should a requirements review produce?
Consider this illustrative requirement: “The unit shall provide an appropriate warning quickly.” A useful review identifies the missing warning definition, trigger and response limit. It preserves the original statement and asks the owner to resolve those values. It should not invent a 100 ms limit because that makes the sentence look measurable.
The verifiability skill connects the wording to observable evidence. The ECSS requirement-review skill additionally requires an inspected, applicable standard edition before making a clause-level finding. Standards are not bundled. Without the controlled text, the result must remain an advisory engineering-quality review.
For verification planning, require a clear distinction between a linked test, an executed test and acceptable evidence for the current configuration. A link labelled “verifies” records scope. It is not a passing result. Our requirements traceability matrix guide explains how those records fit together.
Is local processing private or offline?
The privacy documentation states that the local server has no Arc connection, telemetry or embedded model client. Reads and exports are confined to the selected workspace. The connected assistant and its provider can still process content returned by the tools. Local parsing therefore does not make the complete AI workflow offline.
Use a dedicated folder containing the material intended for the task and assess the host service policies before providing engineering information. Dependency installation can contact package registries. The workspace path guard is also not an operating-system sandbox.
When should you choose Arc Skills?
Choose it when your immediate task is to extract, clean, review or compare engineering information with an AI assistant and retain inspectable files. It can support a workflow around an existing requirements system without replacing that system. The MIT licence permits use and modification under its terms, while the assistant may have its own charges.
Evaluate StrictDoc or Doorstop when your main requirement is a maintained specification in version control. Evaluate a shared requirements platform when you need a multi-user programme record with access controls, formal review and controlled baseline changes. See the seven-tool open source guide and Arc versus open source comparison for those operating-model choices.
Arc Skills and the commercial Arc platform are separate offerings. This repository includes no external application write integration. Applying a proposal elsewhere needs that application's authorised import and review process.
What has been validated?
The repository's validation record reports package checks, local MCP transport checks, deterministic tests and synthetic engineering scenarios. Those are maintainer-reported results. They do not establish certification, production accuracy, installation in every client or performance on your programme's data.
For a practical evaluation, use a small representative specification with known issues. Check source preservation, missed and incorrect findings, row reconciliation, unresolved links and whether another engineer can reproduce the reasoning. Keep formal acceptance with the people responsible for the requirement and evidence.
Frequently asked questions
Is Systems Engineering Skills an open source requirements management tool?
Yes. Systems Engineering Skills, also called Arc Skills, is an MIT-licensed toolkit for requirements-management tasks through an AI assistant. It combines 100 engineering skills with local file tools. It does not include a shared requirements database, formal approval workflow or custom browser application.
Is Arc Skills free to use?
The skills and code are free under the MIT licence, and the toolkit needs no Arc account or model API key of its own. Your chosen AI assistant may charge for access or usage. Licensed engineering standards are not bundled.
Can Arc Skills review requirements against ECSS?
It includes skills for reviewing requirements and verification matrices against ECSS. A clause-level finding needs the applicable, inspected edition and project context. Without that source, the review is advisory and cannot establish compliance.
Does Arc Skills send engineering files to Arc?
The local MCP server has no Arc connection or telemetry. Content returned to the connected assistant may be processed by its AI provider. Local file parsing does not mean offline AI, and uploaded files follow the host service policies.
Can Arc Skills import and export ReqIF?
It reads uncompressed ReqIF and supports reviewed mappings of requirements and links into local proposals. Compressed ReqIF is not supported, and ReqIF export is not included. Available exports are JSON, CSV, XLSX, Markdown and text.
Is Arc Skills the same as the Arc requirements platform?
No. Arc Skills is the open source assistant toolkit. Arc is the separate commercial requirements management and verification platform. The toolkit does not include a connector that applies its proposals to Arc or another requirements application.
Evaluate Arc
Try Arc on a representative engineering workflow
Start with one requirement set and test traceability, change control, review and verification in a private Arc workspace.