feat(skills): add linear-deep-dive skill

[gcko]

Summary

• Adds a new Claude Code skill (/linear-deep-dive) that fetches a Linear issue, classifies it (feature/bug/refactor/investigation), explores the relevant codebase, proposes an approach, and orchestrates the right skill chain to solve it
• Integrates with Linear MCP server, superpowers skills (brainstorming, writing-plans, systematic-debugging, TDD, executing-plans), and Explore agents
• Includes edge case handling for blocked issues, vague specs, full-stack work, and WIP branches

Test plan

• Skill loads correctly via /linear-deep-dive DRC-2893 (verified in session)
• Test with a Feature-labeled issue to verify brainstorming → writing-plans routing
• Test with a Bug-labeled issue to verify systematic-debugging → TDD routing
• Test with an issue that has blockedBy relations

Checklist

• Signed-off-by (DCO)
• No state files committed
• No changes to recce/data/

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds a new Claude skill (/linear-deep-dive) to drive end-to-end analysis of a Linear issue: fetch issue context, classify work type, explore relevant code, produce an approach proposal, and then orchestrate an appropriate skill chain.

Changes:

• Introduces the linear-deep-dive skill definition and workflow documentation.
• Defines default routing from Linear issue classification (feature/bug/refactor/investigation) to skill chains.
• Documents branch handling and several edge-case decision points (blockers, vague specs, full-stack scope, WIP branches).

[gcko]

Code Review — PR #1195

Summary

Adds a new Claude Code skill (/linear-deep-dive) for analyzing Linear issues and orchestrating the appropriate workflow. This is a markdown-only skill definition — no executable code. The skill is well-structured, follows established conventions from the existing skills (claude-code-review, recce-mcp-e2e), and correctly references existing superpowers skills.

Findings

No critical issues found.

Observations (not blocking):

• The skill correctly uses the mcp__claude_ai_Linear__get_issue and mcp__claude_ai_Linear__list_comments tools for fetching issue data
• The classification table (feature/bug/refactor/investigation) with mapped skill chains is well-designed
• Edge cases (blockers, vague issues, WIP branches, full-stack issues) are handled
• The iron rules section enforces the right guardrails (always fetch first, always confirm with user, stay in scope)
• The process flow diagram follows the same digraph convention used in claude-code-review

Verdict

Approved. Clean addition of a single skill file with no impact on existing code.

[kentwelcome]

LGTM