Add a target per agent: Cursor and Codex, then Kiro and Devin #64

Component

Something else: one target folder per agent

Experience

• Operator: in Cursor or Codex, one command installs the plugin, and its skills, sub-agents and rules work through that agent's own support. The same follows for Kiro, then Devin and other agents with a plugin format.
• User (uses what the operator builds): gets the plugin's discipline whichever of those agents the operator uses. The exception is where an agent's question tool does not wait for approval, which is recorded as a gap.
• Owner: adds an agent by adding one target folder: its generator, manifest, hook format and install path, all built from the canonical source.

Proposed behaviour

Add one target at a time: Cursor and Codex first, then Kiro, then other agents as they gain a format. From the spike's runs in #61:

• Codex reads a .claude-plugin marketplace and runs our rule hooks unchanged, but offers no plugin agents. Its target writes the three agents as Codex's own agent files, plus an install step for them.
• Cursor reads the .claude-plugin layout, but takes rules only from a hook that prints them as JSON ({"additional_context": ...}), and ignores agents' tool lists. Its target writes that hook and agents in Cursor's form.
• Kiro needs steering for the rules, Markdown agents with Kiro's tool names, and an installer into ~/.kiro/.
• Devin has a plugin format, which AWS's agent-toolkit-for-aws already targets.

Motivating case

Today the plugin installs only in Claude Code, so an operator in another agent goes without it or switches back.

Done when

Each target installs with one command, and its scenarios from #65 show the plugin's skills, sub-agents and rules working in that agent.

Depends on

#63, once the Claude Code target reproduces today's plugin. This is a stream in .freethinking/strategies/multi-tool-plugin.md. Design with /freethinking:hld .freethinking/strategies/multi-tool-plugin.md agent-targets.