Nexus Agent Kit

Nexus Agent Kit is a plugin-first AI terminal workspace for Codex and Claude Code. It packages specialized subagents, reusable skills, and terminal commands — plus MCP safety conventions — so your AI sessions behave consistently across product specs, architecture reviews, database migrations, cloud cost audits, CI/CD debugging, code review, TODO tracking, and tool-connected work.

Instead of re-explaining how you want the assistant to operate in every new repo or terminal, you install Nexus once and get a shared operating model for engineering work.

Connect

I regularly write about AI, DevOps, Cloud Infrastructure, and Software Engineering. Subscribe to get practical guides, deep dives, and updates.

If this project is useful and you want to support more open-source AI engineering work, you can sponsor it on GitHub.

Quick Start

Nexus is distributed as a git-based plugin marketplace. There is no npm package: the bundled Node scripts are dependency-free and run with plain node (Node 18 or newer), so no npm install is ever needed.

Install In Codex

Install directly from this GitHub repo using the third-party codex-marketplace tool:

npx codex-marketplace add aayushostwal/nexus --plugin --global

Project-scoped install:

npx codex-marketplace add aayushostwal/nexus --plugin --project

This is a direct repo install. Users need the repo slug aayushostwal/nexus; pushing this repo does not by itself make the plugin searchable in a central Codex marketplace index.

After install, Nexus automatically bootstraps global instruction files on first active plugin run:

• Claude runtime: updates ~/.claude/CLAUDE.md
• Codex runtime: updates ~/.codex/AGENT.md

If these files already exist, Nexus appends/updates only its managed block and preserves your existing content.

Install In Claude Code

/plugin marketplace add aayushostwal/nexus
/plugin install nexus@nexus-marketplace
/reload-plugins

After plugin enablement, run node scripts/bootstrap-agent-docs.js once from the cloned or installed plugin directory (plain node, no dependencies to install) to create or refresh the managed skills-first instruction block in your runtime instruction files.

What Nexus Does In Your Terminal

• Turn rough notes into a full PRD, or get an existing PRD critiqued with severity-ranked gaps.
• Plan a production change as an ordered roadmap with verify commands, rollback steps, and an approval gate before any code.
• Get a GO/NO-GO architecture verdict on a deployment, with service boundaries and risk tiers mapped from the codebase.
• Design zero-downtime database migrations and index strategies from real query plans.
• Plan for scale with capacity math and bottleneck-ordered fixes, or design event-driven systems with idempotency and replay handled.
• Audit cloud spend from real billing data and live-verified pricing, then write the Terraform/CDK to fix it.
• Review a branch, PR, or UI for correctness, missing tests, accessibility, and mobile UX issues.
• Debug a failing CI pipeline or a deploy that works locally but fails in GitHub Actions.
• Explore a large codebase token-efficiently, with project memory that persists across sessions.
• Build LLM-powered features end-to-end, and ship docs as a searchable React app.
• Build a daily brief from TODOs and connected work systems.

Skills

Skills are grouped below by role so it is easier to understand what the plugin actually adds to the terminal.

Core Operations

Skill	Category	Purpose
nexus		Shared operating rules, TODO workflows, daily briefs, and MCP safety behavior.

Debugging

Skill	Category	Purpose
nexus-debugging		Investigates failures and regressions with root-cause analysis, narrow fixes, verification, and prevention guidance.
debugging-common		Shared debugging rules, output patterns, and checklists.
debugging-ci-cd		CI/CD-focused debugging playbook.
debugging-codebase		Application and code regression debugging playbook.
debugging-frameworks		Framework and tooling-specific debugging playbook.

Testing

Skill	Category	Purpose
nexus-testing		Identifies why tests fail intermittently and produces targeted fixes rather than retry wrappers.

Reliability And Release

Skill	Category	Purpose
nexus-reliability		Works through an active incident with structured triage, blast radius assessment, and resolution guidance.
nexus-release-readiness		Checks a release candidate against rollback readiness, monitoring coverage, and known risk factors before shipping.

Observability

Skill	Category	Purpose
nexus-observability		Correlates error spikes, latency anomalies, and log patterns across services to surface the most likely failure origin.

Performance

Skill	Category	Purpose
nexus-performance		Entry point for all performance requests — routes to memory, dependency, or profiling tracks based on the symptom.
performance-memory-leak		Instruments a running process, identifies allocation hot paths, and confirms the fix eliminated the leak.
performance-dependency-blast-radius		Maps which modules are affected by a dependency upgrade before the upgrade is applied.

Education And Documentation

Skill	Category	Purpose
nexus-tutorial		Creates executable Jupyter-style tutorials with reproducible setup.
skill-writer		Helps create or improve new SKILL.md workflows.

Content And Social

Skill	Category	Purpose
nexus-shorts		Converts ideas, notes, or technical content into YouTube Shorts scripts.

Token Efficiency

Skill	Category	Purpose
nexus-token-optimizer		Always-on token optimization rules for efficient context gathering, tool usage, and concise responses.

Agents

Agents are autonomous subagents that Claude Code can delegate to. They run in their own context window with a focused system prompt and a restricted tool set.

Every Nexus agent has persistent memory and a domain-grouped display color. user-scoped memory carries your portable preferences across every repo; project-scoped memory learns the conventions of the repo it runs in. Project memory lives in .claude/agent-memory/ — add it to .gitignore if you don't want it committed.

Product (purple)

Agent	Memory	Purpose
prd-writer-critic	user	Writes PRDs from rough notes using a 10-section template with hard KPI/JTBD gates, or critiques existing PRDs with severity-ranked gaps.
roadmap-planner	user	Turns an intended change into a scoping table, ordered steps with verify commands, risks, and rollback plans — hard-stops for approval before any implementation detail.

Design (pink)

Agent	Memory	Purpose
uiux-reviewer	user	Read-only UI/UX review: heuristics, WCAG 2.2 AA accessibility, state coverage, and responsive behavior, with impact-calibrated severity.
mobile-ux-designer	user	Designs and reviews mobile UX across iOS/Android/cross-platform: navigation, touch ergonomics, offline states, permission choreography.

Architecture (blue)

Agent	Memory	Purpose
system-architecture-reviewer	project	Maps service boundaries, coupling, and extraction candidates; issues GO/NO-GO deployment safety verdicts with T1–T5 risk tiers.
scalability-planner	project	Bottleneck-ordered scaling plans with explicit capacity math and tiered triggers — boring solutions first, sharding last.

Data & Events (orange)

Agent	Memory	Purpose
database-architect	project	Schema design, index strategy from real query plans, and zero-downtime expand–contract migrations with per-step rollback.
event-driven-designer	project	Designs and reviews async systems: idempotency, ordering costs, outbox, sagas, DLQ replay, and schema evolution.

Cloud (yellow)

Agent	Memory	Purpose
cloud-cost-optimizer	user	Cost audits from real billing data and live-verified pricing — never from memory. Quick wins, structural changes, and monitoring setup.
iac-engineer	project	Designs and writes Terraform/CDK, audits existing IaC with evidence-cited findings, produces HLDs with verified cost tables.

Code & Docs (green / red)

Agent	Memory	Purpose
codebase-explorer	project	Token-efficient codebase navigation; returns a three-column map (Title, File path, Description) consumable by humans and downstream agents.
code-reviewer	project	Read-only senior reviewer for a PR, branch, or diff. Returns a verdict with file:line findings covering correctness, regressions, security, migration risk, and deploy safety.
docs-app-builder	project	Builds documentation as a React app: Diátaxis-organized sidebar navigation, Mermaid diagrams, reference tables, search, tabs, and dark mode — verified by a passing build and link crawl.

AI (cyan)

Agent	Memory	Purpose
ai-product-engineer	user	Builds LLM-powered features end-to-end: model selection, cost projections, RAG and agent design, evals as the core loop, production failure modes.

Commands

Command	Category	Purpose
/add-todo		Add a classified item to the Nexus TODO system.
/daily-brief		Build a daily brief from TODOs and configured tools.
/review-branch		Review current branch changes with findings first.
/stats		Show token usage and estimated cost stats for Claude and Codex sessions.

Tool Setup Guides

Nexus is designed to work well with MCP-connected tools, but it does not ship credentials or external accounts.

Tool	Guide	What it adds to Nexus
Microsoft / Outlook	Microsoft Graph setup and permission guidance.	Mail and calendar context for daily briefs and workflow assistance.
Slack	Slack MCP connection and token scope guidance.	Workspace search, channel context, thread retrieval, and approved messaging workflows.
Notion	Notion MCP and internal integration setup guidance.	Access to pages, databases, and workspace knowledge.
Jira / Atlassian	Atlassian Rovo MCP and API token setup guidance.	Issue, sprint, board, and project context for engineering workflows.
AWS	AWS profile, role, and least-privilege MCP guidance.	Cloud status, logs, metrics, and infrastructure context with approval gates for changes.

Safety defaults used by the repo:

• Read-only by default.
• Ask before sending messages, email, Jira updates, or cloud changes.
• Keep secrets out of git.
• Prefer scoped credentials, OAuth, or temporary tokens.

How To Extend Nexus

1. Pick the tool you want to connect.
2. Open the matching setup guide in docs/tools/.
3. Configure the MCP connection or provider credentials outside git.
4. Start with read-only access.
5. Only add write permissions when your workflow actually needs them.
6. Keep Nexus approval-first for any external write, send, deploy, or update action.

Usage In Real Life

Title	Labels	Screenshots
Slack TODO Aggregator
Article to Instagram Short Script Conversion

Development

The scripts are dependency-free, so local checks run with the built-in Node test runner:

node --test

Plugin packaging lives in:

• .agents/plugins/marketplace.json
• .codex-plugin/plugin.json
• .claude-plugin/plugin.json
• .claude-plugin/marketplace.json

License

This project is licensed under the MIT License. See LICENSE for the full text.

SEO / GEO Tags

claude code agents, claude code plugin, codex plugin, subagents with memory, ai terminal workflow, developer ai assistant, prd agent, roadmap planning agent, code review agent, architecture review agent, database migration agent, event-driven design agent, cloud cost optimization agent, terraform agent, infrastructure as code review, scalability planning, ui ux review agent, mobile ux design, ai product engineering, rag and llm evals, documentation site generator, ci cd debugging, github actions debugging, mcp tools, model context protocol, slack mcp, notion mcp, jira mcp, aws mcp, ai todo manager, daily brief automation, aws architecture design, cloud cost estimation, devops ai assistant, developer productivity