mirrord Agent Skills

Twelve Agent Skills that close the AI agent feedback loop on Kubernetes: real env vars, real DNS, real network, real traffic, real databases, real Kafka, real Temporal task queues, real preview environments, multi-service local sessions, real failure conditions, the header propagation that lets a session follow a request across services and queues, and the discovery of what your cluster already propagates so a filter can use it. Built by MetalBear to be used with mirrord.

AI coding agents work from what's in their context window. Your Kubernetes cluster is full of state that isn't. These skills teach your agent how and when to use mirrord so the code it writes against your live infrastructure stops being informed guessing.

Installation

Claude Code

/plugin marketplace add metalbear-co/skills
/plugin install mirrord@mirrord-skills

Codex

codex plugin marketplace add metalbear-co/skills

Then install the mirrord plugin from /plugins. All twelve skills come with it.

OpenCode

OpenCode loads skills natively, so there's nothing to register — the skills just need to be on disk:

curl -fsSL https://raw.githubusercontent.com/metalbear-co/skills/main/install.sh | sh

That writes the twelve skill folders into ~/.config/opencode/skills/. Restart OpenCode and they're available. Re-run it any time to update — it replaces the mirrord-* folders and leaves your other skills alone.

The same script serves any agent that reads a skills directory:

sh install.sh --agent codex     # ~/.agents/skills
sh install.sh --agent claude    # ~/.claude/skills
sh install.sh --dest <dir>      # anywhere else

Cursor, Gemini, or any other Agent Skills-compatible agent

npx skills add metalbear-co/skills

GitHub Copilot, Cline, and other rules-file agents

The ports/ directory carries the same core content as drop-in rules files: copy ports/github-copilot/copilot-instructions.md into your repo's .github/, or ports/cline/.clinerules into your repo root. Want another format? Open an issue with the target you'd like next.

The twelve skills

Skill	What it does
mirrord-quickstart	Zero-to-first-session: install mirrord, find a target, run the first session.
mirrord-config	Generate and validate mirrord.json for any workflow (steal, mirror, env injection, file system hooks).
mirrord-operator	Install and configure the mirrord Operator for team-scale concurrent shared-cluster use, including agent-started trials on clusters with no license yet.
mirrord-ci	Wire mirrord into CI pipelines so integration tests hit real cluster services.
mirrord-db-branching	Per-developer copy-on-write database branches off your staging DB.
mirrord-kafka	Kafka queue splitting so each developer consumes a private slice of a real topic.
mirrord-temporal	Temporal task queue splitting so each developer's local worker gets only the workflow and activity tasks matching their filter.
mirrord-prev-env	Preview environments: deploy a built image as an isolated, traffic-filtered pod in a shared cluster — ad hoc or per-PR in CI.
mirrord-up

Example prompts

Once installed, your agent activates the right skill based on the prompt:

• "I'm new to mirrord, help me run my Node app against my staging cluster."
• "Steal traffic from pod/api-server, but only requests carrying my baggage header so I don't break anyone else's session."
• "Install the operator on our EKS cluster and configure RBAC so only the dev group can use it."
• "Set up GitHub Actions to run our integration tests with mirrord against the staging cluster."
• "Give me an isolated DB branch off the staging Postgres for this feature."
• "Set up queue splitting on the orders.created topic for my local consumer."
• "Route only Temporal workflows whose ID starts with test-local- to the worker on my laptop."
• "Spin up a per-PR preview environment in GitHub Actions so the team can review my change against real traffic."
• "Generate a mirrord-up.yaml so I can debug my auth and dashboard services together."
• "Make 30% of my service's calls to the payments API time out, so I can test our retry logic."
• "Make the mirrord session header follow requests through our Kafka and SQS consumers."
• "Which header or JWT claim can I filter on so I only get my own requests across all our services?"

Trusted by

mirrord runs across teams like monday.com (350+ engineers on a single shared staging cluster), SurveyMonkey, CoLab, Cadence, Daylight Security, and Zooplus. Daylight's team cut their edit-test cycle from 5–8 minutes to about 5 seconds after pairing Cursor with mirrord.

Data and external services

The skills are instructions only. They contain no code that runs on install. The agent runs mirrord, kubectl, and helm against the cluster in your kubeconfig.

One skill can send data to a service outside your cluster. When a cluster has no mirrord for Teams license, mirrord-operator can start a seven-day trial, but only after you agree. It sends these values to MetalBear (app.metalbear.com), through mirrord operator install or the /api/v1/agent/signup endpoint:

• The name of the agent, for example claude-code
• Your email address, if you give it
• A hint that identifies the cluster: the UID of its default namespace, or a name you choose

MetalBear keeps this data as its privacy policy describes. The mirrord CLI and the Operator also send usage telemetry to MetalBear. The mirrord documentation tells you how to turn it off.

Skill structure

Each skill is a folder following the Agent Skills format:

• SKILL.md - Instructions the agent loads on demand
• scripts/ - Helper scripts for automation (optional)
• references/ - Supporting documentation (optional)

Learn more

• mirrord for AI Agents - product page with installation guides for every supported agent
• mirrord documentation

Contributing

PRs welcome. Open an issue if you'd like to see a skill added or improved. See CONTRIBUTING.md for the json/yaml fence tag convention.