Add Cursor harness support for `skore agent`

[louiscrc]

Summary

Request first-class support for Cursor in the skore agent command.

Today, skore-cli already detects Cursor via the CURSOR_AGENT environment variable and adapts behavior for skills installation and in-agent workflows. However, the skore agent subcommand only lists Claude, OpenCode, Pi, and GitHub Copilot as supported harnesses.

When running skore agent from a Cursor session, the docs say it auto-selects the detected agent's harness and skips the launch step — but Cursor is not listed among the harnesses that can be configured or launched from the CLI.

Motivation

Cursor is a widely used coding agent, and many Skore users (including workshop participants) work inside Cursor. Being able to run skore agent end-to-end with Cursor would allow:

• Connecting a project to the Skore Hub agent from Cursor
• Writing the appropriate harness configuration for Cursor (rules, skills, .skore setup)
• A consistent onboarding flow: skore agent → workspace selection → harness config → ready to work

This would mirror the experience already available for Claude Code, OpenCode, Pi, and GitHub Copilot.

Expected behavior

Something along the lines of:

skore agent
skore agent --harness cursor

When Cursor is detected (CURSOR_AGENT set), skore agent would:

1. Log in / pick workspace / create API key (same as other harnesses)
2. Write Cursor-specific harness configuration (e.g. project rules, skill paths under .cursor/skills/ or equivalent)
3. Either launch Cursor where applicable, or skip launch when already running inside Cursor (consistent with current in-agent detection behavior)

Current state (from README)

Area	Cursor support
skore skills	✅ (--agent cursor, auto-detection via CURSOR_AGENT)
skore agent harness list	❌ Not listed
Agent detection table	✅ CURSOR_AGENT

Context

Related: #1 (env var auto-detection for agents)

Happy to help test or provide feedback on a Cursor harness implementation.