Open tools. Shared standards.
Give your agents a shared way to work.
Turn individual prompts into reusable capabilities. AI Cortex brings skills, contracts and quality rules to software delivery and project governance.
Open source under MIT. Built for the tools you already use.
One library. Clear boundaries.
Capabilities you can reuse across projects.
Start with the work in front of you
A useful next step.
A repeatable way to take it.
Choose a task. Find a capability with a defined purpose, process and output. Combine capabilities when the work needs a longer chain.
Delivery & reviewMake the next change reviewable.
Review a diff, check engineering quality, prepare a commit or assemble a release package. Keep criteria explicit and the resulting evidence available.
orchestrate-code-reviewRun the engineering review chain ↗commit-workCreate a scoped, reviewable commit ↗prepare-releaseBuild and validate a release package ↗
Governance & planningConnect the plan to the purpose.
Capture requirements, define strategic direction and maintain a roadmap grounded in project evidence. Keep priorities and dependencies traceable.
capture-work-itemsTurn an idea into a persistent work item ↗define-roadmapShape and populate a roadmap ↗plan-nextFind the next governance action ↗
Research & opportunityBring evidence to the decision.
Investigate an open question, compare a market or assess a product opportunity. Separate source-backed findings from the decision they inform.
deep-researchInvestigate and corroborate a question ↗competitive-researchCompare competitors with dated evidence ↗product-opportunity-analysisAssemble an opportunity package ↗
The foundation
Good instructions have boundaries.
AI Cortex separates what an agent can do from the contracts and constraints it works within. The result is a library you can compose and maintain.
- CapabilitySkill
What can the agent complete?
A single agent follows an objective and execution process to produce a defined output.
Explore skills- ContractSpec
What does this thing look like?
The structural and behavioral contract for an artifact, data model or interface.
Explore specs- CoordinationProtocol
How do the parties interact?
The steps, states and message sequence that coordinate two or more roles.
Explore protocols- ConstraintRule
What must hold true?
An independently verifiable obligation or prohibition, shared by the work and its review.
Explore rules
How it fits together
A shared standard.
From planning to delivery.
An example path through the library. Start with the capability you need; a full workflow is optional.
- Frame the work
Capture the need and retain its purpose, acceptance criteria and strategic connection.
Capture a work item - Check the evidence
Review the implementation against engineering rules and check alignment with the approved requirement.
Understand the quality gates - Prepare the handoff
Package changes with the validation and release information the next person needs.
Prepare a release
Make it part of your workflow
One library.
Your environment.
Clone the library and run the installer. It links skills into supported agent environments and syncs user-scoped rules for detected IDEs.
Read the full installation guidemkdir -p ~/.local/share
git clone --depth 1 https://github.com/nesnilnehc/ai-cortex.git ~/.local/share/ai-cortex
~/.local/share/ai-cortex/bin/cortex install
Review the commands before running them. The installer reports detected IDEs and any PATH setup needed.
Adopting project quality rules? Read the configuration guideBefore you start
A few useful answers.
Is AI Cortex an agent or an application?
It is an asset library for agents. It supplies callable skills, structural specs, interaction protocols and quality rules. Your existing agent runs the work.
Do I need to adopt everything?
No. Start with a skill that fits a current task. Orchestrator skills combine smaller capabilities for longer workflows. Project-specific governance needs its own explicit adoption. See the adoption guide.
What gets installed?
The installer links skills into ~/.agents/skills and supported IDE locations. User-scoped rules are synced for detected IDEs. Specs, protocols and project-scoped rules stay in the canonical clone for agents to read when needed. See supported installation behavior.
Can I inspect and adapt the library?
Yes. The project is open source under the MIT license. Read the assets, reuse them across projects and contribute improvements through the repository. Read the contribution guide.