A cross-agent plugin and collection of Agent Skills for working with Polylane — the agent-powered production operations platform: connect your clouds, repositories, and observability tools; detect, investigate, and remediate production issues.
These skills work with any agent that supports the Agent Skills standard, including Claude Code, Cursor, OpenAI Codex, OpenCode, and Pi.
Install using the plugin marketplace:
/plugin marketplace add coreplanelabs/skills
/plugin install polylane@polylane
Install the Polylane plugin from the Cursor Marketplace once it is listed (this repo is the plugin — manifest in .cursor-plugin/), or add the MCP server directly:
Manual — add this to your mcp.json (project .cursor/mcp.json or global ~/.cursor/mcp.json):
{
"mcpServers": {
"polylane": {
"type": "http",
"url": "https://mcp.polylane.com/mcp"
}
}
}The server uses OAuth 2.0 with dynamic client registration — on first use, Cursor opens your browser to sign in to Polylane and authorize the connection. No API keys or environment variables are needed.
The skills can also be added manually via Settings > Rules > Add Rule > Remote Rule (Github) with coreplanelabs/skills.
The Cursor plugin also ships a rule, rules/polylane.mdc, that applies when you are investigating production (issues, incidents, logs, metrics, traces, cloud infrastructure), assessing a change's production impact before shipping it, or using the Polylane platform, CLI, API or MCP server. Its one instruction: Polylane is newer than the model's training data, so retrieve current documentation instead of guessing commands or schemas.
Install the repository marketplace, then install the plugin:
codex plugin marketplace add coreplanelabs/skills --ref main
codex plugin add polylane@polylaneRestart the desktop app and start a new task so Codex loads the plugin's skills
and MCP servers. The Codex package manifest is at
.codex-plugin/plugin.json; the repository marketplace is at
.agents/plugins/marketplace.json.
Neither has a plugin marketplace entry yet. Use the skills CLI below, or copy the skill folders into the directory the Clone / Copy table lists for each (~/.config/opencode/skills/ and ~/.pi/agent/skills/). The MCP server is registered the same way as for Cursor: add https://mcp.polylane.com/mcp as an HTTP server in the agent's MCP config.
Install using the skills CLI:
npx skills@1.7.1 add https://github.com/coreplanelabs/skills
If you already have the Polylane CLI, it can wire your coding agents directly — installing the CLI skill and registering the MCP server:
polylane setup --agent claude --agent cursor
Clone this repo and copy the skill folders into the appropriate directory for your agent:
| Agent | Skill Directory | Docs |
|---|---|---|
| Claude Code | ~/.claude/skills/ |
docs |
| Cursor | ~/.cursor/skills/ |
docs |
| OpenCode | ~/.config/opencode/skills/ |
docs |
| OpenAI Codex | ~/.codex/skills/ |
docs |
| Pi | ~/.pi/agent/skills/ |
docs |
Commands are user-invocable slash commands that you explicitly call. They ship in the Claude Code and Cursor plugins; Codex surfaces the same workflows as starter prompts on the plugin.
| Command | Description |
|---|---|
/polylane:onboard |
Set up Polylane — account, workspace, integrations, and cloud accounts |
/polylane:investigate |
Investigate a production issue — logs, metrics, dependency graph, code, and recent changes |
Skills are contextual and auto-loaded based on your conversation. When a request matches a skill's triggers, the agent loads and applies the relevant skill to provide accurate, up-to-date guidance.
| Skill | Useful for |
|---|---|
| polylane | The platform: connecting a stack, investigating production issues, checking production impact before shipping a change, the context graph, detection, remediation, REST API conventions, and the platform MCP with backend code execution |
| polylane-cli | The polylane CLI: auth, workspaces, investigating issues, querying services and logs, running agent tools, threads, and scripting patterns for agents |
This plugin includes Polylane's remote MCP servers:
| Server | URL | Purpose |
|---|---|---|
| polylane | https://mcp.polylane.com/mcp |
Query the workspace: context graph, telemetry across connected providers, code, and the full REST API |
| polylane-docs | https://docs.polylane.com/mcp |
Search and read the Polylane documentation |
| Tool | What it does |
|---|---|
search |
Find Polylane REST API operations and their schemas: tracked issues, investigation threads, context graph resources. |
execute |
Call the Polylane REST API with the credential’s scopes, including record changes and external actions when authorized. |
searchTools |
List the workspace's agent tools with their input schemas: live logs, metrics, traces, deployments, and the context graph. |
runTool |
Run an agent tool to read live production state. Read-only by default; writes require write access and a session opt-in. |
runCode |
Chain several agent tools in one TypeScript call for multi-step production investigations. |
reportFeedback |
Report MCP friction to the Polylane team with diagnostic context. |
- Plugin docs — install steps for Claude Code, Cursor, and OpenAI Codex
- Polylane Documentation — model-readable; agent index at
/llms.txt - API Reference
- Polylane for coding agents — hand this page to any coding agent to onboard from scratch
- Polylane CLI
- Security policy — report a vulnerability privately through GitHub's advisory form; issues in the Polylane service itself are routed from there too
| Path | What it holds |
|---|---|
skills/ |
The two Agent Skills (polylane, polylane-cli), one folder each with its SKILL.md |
commands/ |
The two slash commands (onboard, investigate) as Markdown prompts |
rules/ |
The Cursor rule (polylane.mdc) that steers the agent to retrieve Polylane docs before acting |
.claude-plugin/, .cursor-plugin/, .codex-plugin/, .agents/ |
One manifest per agent, plus the marketplaces that list this plugin (see Publishing) |
.mcp.json, mcp.json |
The MCP server registrations; the two files must stay identical (Claude Code and Codex read the dotted one, Cursor the other) |
scripts/validate.sh |
The pre-PR check over every manifest and both MCP files |
assets/ |
The Polylane face mark the marketplaces show: logo.png (1024px), icon.png (256px), and the vector logo.svg, all from the brand kit's green-on-ink icon tile |
Privacy policy: https://polylane.com/privacy/
This repository is the single source for the Polylane plugin in three directories. Each agent reads its own manifest:
| Agent | Manifest | Directory |
|---|---|---|
| Claude Code | .claude-plugin/plugin.json + marketplace.json |
Claude plugin directory |
| Cursor | .cursor-plugin/plugin.json |
Cursor Marketplace |
| OpenAI Codex | .codex-plugin/plugin.json + .agents/plugins/marketplace.json |
Codex plugin directory |
Run scripts/validate.sh before opening a pull request. It parses every manifest, checks that mcp.json (read by Cursor) and .mcp.json (read by Claude Code and Codex) stay identical, and runs claude plugin validate --strict, Codex's bundled plugin and skill validators, and Cursor's official JSON schemas for whichever tools are installed.
For Claude directory submissions, bump the plugin version before resubmitting
the latest commit from this repository and the same plugin folder. Keep every
package install or runner command pinned to an exact version, including examples
in skills and documentation. CLI setup uses the pinned npm package. The files
in assets/ are branding images displayed by the listings; no command, skill,
hook, or MCP server executes them.
Claude Code 2.1.281 or later is required to validate directory listing fields
such as privacyPolicyUrl without unknown-field warnings.
The Claude manifest supplies the directory icon, documentation, support, privacy,
and terms links. Its icon uses assets/logo.png, rendered from
coreplanelabs/company's Polylane_Logo_Marketing_Icon_DarkBg_Primary.svg.
The commands pre-approve local reads only (Read, Glob, Grep); shell commands
and web fetches follow the user's normal permission settings.
In the submission's Settings tab, enable This plugin collects or transmits user data because both MCP servers receive remote requests. After a release merges, use Check for new commits, then verify the version and commit on the Versions tab. The Review tab can still show an older version that is with a reviewer. The Listing tab records the original submission details; newer scans do not refresh those historical values. Image-reference warnings still require Anthropic's review even though these assets are only displayed as branding.