MCP integration with task tracker

Hooks the team's task tracker (Linear, Jira, Kaiten, Asana) into the coding agent via MCP. Workflow becomes: "give the agent the ticket number → it pulls context (description, comments, linked specs, related code) → it edits → commits → opens MR linked back to the ticket." Reported 3–5× speedup on typical tickets.

Most of the latency in shipping a small fix isn't the code — it's the "open tracker, read context, find the file, open editor, switch back, paste context" round-trip. MCP collapses that into one command.

1. Install the tracker's MCP server (most have one; if not, write a tiny one — see #158).
2. Add to AGENTS.md: "When given a ticket ID like ABC-123, call the tracker MCP, fetch the ticket + all comments + linked tickets, and load context before editing."
3. Add a slash-command or skill do-ticket that wraps the flow: fetch → plan → edit → run tests → commit on a branch feature/ABC-123-{slug} → open MR with the ticket link.
4. Configure rules: do not close the ticket; humans review the MR and close.
5. Track wall-clock per-ticket time before/after.

• Stale ticket context is a real problem. If a ticket has 30 old comments and one new spec, the agent over-weights the noise. Have the skill read the most-recent N comments first and ask if it should expand.
• Tracker permissions: the agent should read everything the engineer can but not write more than the engineer typically writes.