v0.3.0 — Agent skill + clinear-mcp server

Added

• Agent skill bundle at skills/clinear/ — a Swarm/Claude-style skill
  (SKILL.md frontmatter + references/*.md) that teaches AI agents both the
  mechanics of the CLI and the behavior of working through Linear (when to
  search before create, when to use --output json, how to chain commands,
  safety rules, anti-patterns). Install with bash skills/install.sh.
• clinear-mcp MCP server — optional Model Context Protocol server
  exposing:
  • 1 tool clinear_guide(topic) returning structured teaching content
    (overview / commands / workflows / filters / output-formats
    / examples).
  • 7 read-only resources: clinear://me, clinear://issue/{id},
    clinear://team/{key}, clinear://project/{id_or_slug},
    clinear://cycle/current/{team_key}, clinear://issues/mine,
    clinear://issues/team/{team_key}.
  • 6 prompt templates: triage, daily_standup, create_from_error,
    hand_off, cycle_review, issue_investigate.
• By design no mutation tools on the MCP server. Mutations happen via
  the clinear CLI in a shell — every tool/prompt response carries a
  reminder reinforcing this rule.
• clinear/skill_content/ — canonical markdown source bundled inside the
  wheel via [tool.hatch.build.targets.wheel.force-include]. Single source
  of truth for both the skill bundle and the MCP server.
• New optional dependency extra: pip install 'clinear[mcp]' pulls in
  mcp>=1.12.0. Core install stays untouched for users who don't need MCP.
• New console script clinear-mcp (entry point of the MCP server).

Changed

• Bumped version to 0.3.0.
• Updated User-Agent string emitted by LinearClient to clinear/0.3.0.

Notes

• The MCP code is lazy-imported. Users who install clinear without the
  [mcp] extra are not affected; the clinear-mcp script prints a friendly
  install hint and exits 2 if invoked without the dep.
• ALL MCP server logging is routed to stderr to keep stdio JSON-RPC clean.

---