Skip to content
v0.0.2MIT

Cluster Linear issues into topic groups, flag duplicates and overlaps, and optionally write the groups back as labels or parent issues.

linear

One canonical skill that pulls issues from Linear, clusters them into topic groups, flags duplicates and overlaps, and (only on request) writes the grouping back as topic labels or parent issues. Packaged for Claude and Codex/ChatGPT.

Layout

.claude-plugin/plugin.json
plugin.json
skills/topic-grouper/SKILL.md
skills/topic-grouper/agents/openai.yaml

This package lives inside the agent-plugins master repository, which owns both marketplace catalogs. Its marketplace identity is danielh-official-plugins. See the master README for installation and full validation.

Requirements

  • The host agent must have the Linear MCP server (or an equivalent Linear app connection) installed and authenticated. This plugin ships only a skill, no Linear tools. Connect Linear in each client separately.
  • Write-back (labels or parent issues) needs a Linear connection that can create labels and edit issues. The skill never writes without an explicit yes.

Editing and validation

  1. Edit the canonical skill only with explicit approval of wording changes.
  2. Bump version in both plugin manifests.
  3. Run claude plugin validate ..
  4. From the master root, run every step of .github/workflows/validate.yml; the root scripts/validate.py checks this package.

Local use

For Claude Code, load this folder with claude --plugin-dir .. For a marketplace installation in Claude Code or Codex, add the master root as described in its README.

Replacing the standalone Claude skill

Disable the standalone linear-topic-grouper skill under Customize > Skills before enabling the plugin, so two copies don't compete. Keep it disabled rather than deleted until the plugin works in a fresh conversation.

Platform limitations

The skill was written for claude.ai. It names ask_user_input_v0 for the scoping question and tool_search for loading Linear tools; other clients lack those exact tools and fall back to asking in plain text and to their own tool loading. The wording is kept as written; changing it needs approval.