TalkOps DevOps Plugins Marketplace
The TalkOps DevOps Plugins Marketplace (talkops-ai/devops-plugins) is an open-source catalog of 40 production-grade plugins shipping specialist sub-agents, domain skills, multi-phase slash-command playbooks, deterministic PreToolUse / PostToolUse safety hooks, and pre-configured Model Context Protocol (MCP) servers for AWS architecture, cloud engineering, FinOps, security, and SRE operations.
Every plugin in the marketplace ships simultaneously in three layouts—Claude Code (.claude-plugin/), OpenAI Codex (.codex-plugin/ + codex/agents/), and the portable Agent Plugins 1.0.0 standard (plugin.json, skills/, mcp.json) supported natively by TalkOps Framework (talkops), OpsCode (ops), Kubernetes Agent (k8s-autopilot), Cursor, GitHub Copilot, Gemini CLI / Antigravity, and Windsurf.
Self-contained Principal-level AWS specialist sub-agents—each with its own isolated context, scoped MCP servers, domain skills, and deterministic guardrails.
245 domain skills and 109 multi-phase slash-command playbooks that attach directly to your main agent without a sub-agent hand-off, plus official HashiCorp Terraform & Crossplane plugins.
Install plugins in 60 seconds across TalkOps CLI, OpsCode, Claude Code, OpenAI Codex, and any Agent Plugins 1.0.0 host.
Defense-in-depth protection combining read-only MCP server defaults, deterministic secret-safety.py and aws-mutation-gate.py hooks, and AWS IAM condition keys.
Marketplace at a glance
The marketplace organizes 40 plugins across three architectural categories so you can choose between isolated sub-agent delegation and direct main-agent capability binding:
| Category | Count | Contents | How It Binds to Your Runtime | When to Choose |
|---|---|---|---|---|
| Agent Plugins | 17 | 1 named Principal AWS specialist per plugin (agents/<name>.md + codex/agents/<name>.toml), up to 33 skills, slash commands, hooks, and scoped MCP servers | Spawns as an isolated dynamic sub-agent with its own tools: allowlist (mcp__plugin_<plugin>_<server>__*) and memory context | You want an end-to-end specialist (e.g., aws-sre-agent, aws-iac-engineer, aws-finops-engineer) to run deep investigations without bloating your main thread |
| Vertical Plugins | 21 | 245 skills total, 109 slash commands total, secret-safety.py & aws-mutation-gate.py hooks, and domain MCP servers | Attaches directly to your main agent—no sub-agent hand-off required | You want deep AWS domain skills and /<plugin>:<command> playbooks (e.g., aws-networking, aws-serverless, aws-databases) active directly inside your primary session |
| Partner-Built Plugins | 2 | Official vendor & ecosystem bundles: terraform (HashiCorp, 16 skills) and crossplane-aws-engineer (TalkOps) | Attaches directly to the main agent (or spawns a specialist sub-agent for Crossplane) | You work with cloud-agnostic HashiCorp Terraform/Plugin Framework or Kubernetes Crossplane AWS compositions |
Architecture: Agent plugins vs. vertical plugins
Why two AWS plugin models?
- Context isolation with Agent Plugins: Running an incident investigation across CloudWatch Logs, X-Ray traces, Prometheus metrics, and CloudTrail events generates thousands of lines of tool output. Delegating to
aws-sre-agentkeeps that telemetry inside an isolated sub-agent context and returns a concise Root Cause Analysis to your main session. - Zero-handoff speed with Vertical Plugins: When you are actively pair-programming a VPC module or debugging a Lambda function in your main thread, installing
aws-networkingoraws-serverlessgives your primary agent immediate access to domain skills and multi-phase slash commands like/aws-networking:trace-connectivityor/aws-serverless:optimize-cold-startswithout spawning a sub-agent.
Multi-host compatibility matrix
Every plugin is validated by scripts/validate_plugins.py to guarantee compatibility across TalkOps, Claude Code, OpenAI Codex, and any host implementing the Agent Plugins 1.0.0 specification:
| Capability | TalkOps (talkops) & OpsCode (ops) | Claude Code (claude) | OpenAI Codex (codex) | Other Agent Plugins 1.0.0 Hosts |
|---|---|---|---|---|
Domain Skills (skills/*/SKILL.md) | ✅ Automatic 7-tier progressive disclosure | ✅ Auto-discovered + /<plugin>:<skill> | ✅ Auto-discovered | ✅ Auto-discovered (plugin.json) |
| Bundled MCP Servers | ✅ Loaded via .mcp.json / mcp.json | ✅ Scoped via .mcp.json | ✅ Loaded via portable mcp.json | ✅ Loaded via portable mcp.json |
| Specialist Sub-Agents | ✅ Dynamic subagent loader (SubagentMemoryStore) | ✅ Native agents/<name>.md isolation | ✅ Copy codex/agents/<name>.toml to ~/.codex/agents/ | Host-dependent |
| Deterministic Safety Hooks | ✅ PreToolUse / PostToolUse lifecycle hooks | ✅ Native (deny / ask decisions) | ✅ Supported after hook trust approval | Host-dependent |
Slash Commands (commands/*.md) | ✅ Custom command runner | ✅ /<plugin>:<command> | Covered via equivalent domain skills | Host-dependent |
Quick install
Install the marketplace catalog and add specialist agents or vertical capability bundles in seconds:
# TalkOps Framework CLI
talkops plugin marketplace add talkops-ai/devops-plugins
talkops plugin install aws-sre-agent
talkops plugin install aws-networking
# OpsCode CLI
ops plugin marketplace add talkops-ai/devops-plugins
ops plugin install aws-iac-engineer@talkops-devops-plugins
# Claude Code
claude plugin marketplace add talkops-ai/devops-plugins
claude plugin install aws-sre-agent@talkops-devops-plugins
claude plugin install aws-serverless@talkops-devops-plugins
# OpenAI Codex
codex plugin marketplace add talkops-ai/devops-plugins
Repository structure
The talkops-ai/devops-plugins repository is organized as a self-contained, dependency-free monorepo:
.claude-plugin/marketplace.json # Claude Code & TalkOps marketplace catalog (name: talkops-devops-plugins)
.agents/plugins/marketplace.json # OpenAI Codex marketplace catalog
plugins/
agent-plugins/<agent>/ # 17 specialist sub-agent plugins:
# .claude-plugin/plugin.json · agents/<agent>.md · skills/ · commands/
# .mcp.json · hooks/ · plugin.json · mcp.json (portable)
# .codex-plugin/plugin.json · codex/agents/<agent>.toml
vertical-plugins/<domain>/ # 21 main-agent vertical bundles:
# .claude-plugin/plugin.json · skills/ · commands/ · .mcp.json
# hooks/ · CONNECTORS.md · plugin.json · mcp.json · .codex-plugin/plugin.json
partner-built/<plugin>/ # 2 vendor/ecosystem plugins (terraform, crossplane-aws-engineer)
src/
hooks/aws-mutation-gate.py # Canonical TalkOps AWS mutation safety gate
scripts/
_plugin_lib.py # Shared dependency-free frontmatter and schema helpers
validate_plugins.py # Structural, binding, portable-format, and Codex validation across all 40 plugins
docs/
authoring-agent-plugins.md # Standards for authoring and maintaining agent, vertical, and partner plugins
iam-guardrails.md # AWS IAM policies and MCP condition keys backing read-only & approval rules
mcp-server-coverage.md # Complete 62-server awslabs/mcp and ecosystem coverage matrix
Explore the documentation
Prerequisites, credential preflight, and step-by-step setup for TalkOps, OpsCode, Claude Code, Codex, and Cursor.
Complete directory of all 17 Principal AWS sub-agent plugins across Build, Platform, SRE, Security, FinOps, Data/AI, and Migration.
All 21 AWS vertical domain bundles (245 skills, 109 slash commands) plus HashiCorp Terraform and Crossplane AWS plugins.
Full 62-server AWS MCP coverage matrix, connection-bound connectors, and the contributor guide for authoring & validating plugins.