feat: integrate Vale with adopted Simple English checks

[N4M3Z]

Summary

Use Vale as the shared prose-lint engine for local prek checks and the GitHub quality pipeline. Adapt useful MIT-licensed rule data, fixtures, and integration lessons from jyooi/agent-simple-english@0.3.1 instead of adopting its runtime or hook system.

The current SimplifiedTechnicalEnglish skill keeps its stdlib-only Python checker for on-demand writing and benchmark scoring. Vale is a later repository-enforcement layer, not a replacement required by the current STE adoption.

Scope

• Pin Vale and verify its release artifact before CI execution.
• Install one reviewed rule set where Vale discovers it for local and CI runs.
• Add a prek commit-stage hook and run the same hook in the GitHub prek pipeline.
• Adapt mechanical checks and fixtures from agent-simple-english@0.3.1: contractions, phrasal verbs, semicolons, terminology, sentence length, verb forms, hedging, and marketing language.
• Study its PreToolUse gate and feedback formatting for useful behavior, without copying its Claude Code hook runtime.
• Preserve source, pinned revision, license, and transformed-file provenance for adopted rule data.
• Measure false positives before making every rule blocking.
• Report only that the configured STE-oriented checks passed.

Non-goals

• Do not add SessionStart, Stop, UserPromptSubmit, reply-rewrite, or per-session Claude Code hooks.
• Do not depend on Bun or the agent-simple-english runtime.
• Do not retain prose, paths, or session observations as agent-simple-english does by default.
• Do not copy the official ASD-STE100 dictionary.
• Do not claim ASD endorsement, certification, or complete ASD-STE100 compliance.

Acceptance

• Local prek and GitHub quality use the same pinned Vale binary and rule files.
• Linting requires no network access after tool installation.
• Code blocks, identifiers, frontmatter, and quoted output are excluded.
• The SimplifiedTechnicalEnglish skill remains responsible for semantic checks such as fact, modality, and scope preservation.
• The existing Python checker remains available for on-demand skill runs and post-generation benchmark metrics.