feat: aspens doc improve — update specific skills interactively

[mvoutov]

Summary

New command: aspens doc improve [domain...] — lets users update specific skills without re-running the full doc init pipeline.

Use cases

• Retry failed domains: aspens doc improve performance-tracking
• Refresh stale skills: aspens doc improve billing auth (after major refactor)
• Improve all: aspens doc improve (interactive picker)
• Re-discover domains: aspens doc improve --rediscover (re-runs discovery, then lets you pick which to regenerate)

Behavior

1. Reads existing skills from .claude/skills/
2. If domains specified on CLI, generates just those
3. If no domains specified, shows interactive picker with existing skill names
4. Uses improve strategy by default (reads current skill, updates based on code changes)
5. Skips base skill and CLAUDE.md unless explicitly requested (--include-base)

How it differs from doc init

	doc init	doc improve
Discovery	Always runs (2 agents)	Skips (uses existing skills)
Strategy prompt	Yes	Defaults to improve
Base skill	Always generates	Skips unless --include-base
CLAUDE.md	Always generates	Skips unless --include-base
Domain picker	Shows all discovered	Shows only existing skills
Hooks	Installs/updates	Skips (already installed)

This would replace the current awkward doc init --mode chunked --domains "x" workaround.