Improve codebase-readiness skill from skill-creator review

[dgalarza]

Summary

• Fix scoring bug: Phase 4 report template had hardcoded weights that didn't match either the dynamic or static weight tables from Phase 3. Report now uses placeholders with explicit instruction to apply the correct language-tier weights.
• Extract recon script: Moved 35 lines of inline bash commands into scripts/recon.sh — more reliable, token-efficient, and fixes false-negative output for CI/CD, linting, and README detection.
• Move report template to assets: 130-line markdown template extracted to assets/report-template.md, keeping SKILL.md focused on procedural instructions.
• Add unsupported language fallback: Skill now gracefully handles codebases in Rust, C#, Kotlin, etc. instead of failing silently.
• Writing style + description cleanup: Converted to imperative form per skill-creator guidelines, tightened frontmatter description.
• Remove authoring notes: Deleted "thinner reference file with room for expansion" meta-notes from Go and Java language files.
• Agent language scoping: Each agent prompt now instructs to reference only relevant sections of the language file.

Test plan

• Packaging validation passes (package_skill.py)
• Recon script tested against CreatorSignal (Ruby/Rails)
• Recon script tested against tracewell.ai (Ruby + JS multi-language)
• Full end-to-end assessment on a real codebase to verify weight tables apply correctly
• Test unsupported language fallback with a Rust or C# project