[Epic] Skills: Execution isolation, model routing & enhanced capabilities

Skills Epic

This epic tracks the implementation of advanced skill capabilities to achieve feature parity with Claude Code's skill system. The key theme is subagent execution — skills should be able to run via the delegation system with configurable agent types, models, and enhanced capabilities.

Sub-Issues

• #2042 — Run skills via subagent delegation for isolated execution
• #2043 — Support agent_type in skill frontmatter for subagent execution
• #2044 — Model override per skill
• #2045 — Shell command execution in frontmatter for context injection
• #2046 — Extended thinking mode for skill execution

Implementation Order

P0

1. #2042 Subagent execution for skills (foundation — wires skills to delegation system)
2. #2043 Agent type selection (depends on #2042)
3. #2044 Model override (independent, can parallel with #2043)

P1

4. #2046 Extended thinking (independent)
5. #2045 Shell commands (independent)

Start with sub-issue #2042, which establishes isolated skill execution through the delegation system, then review the dependent #2043 and independent #2044 and #2046 requirements. The epic is done when the unchecked sub-issues provide subagent execution, agent-type and model configuration, and extended-thinking support for skills.