0.5.0

[cleverhoods]

0.5.0 — AI instruction diagnostics engine

16 commits | 736 unit tests passing | 101 integration tests passing | mypy clean | ruff clean

What changed

• Self-contained install — rules, schemas, agent configs, and ONNX embedding model (all-MiniLM-L6-v2) bundled in the 80 MB wheel. pip install reporails-cli works
  immediately, no ails install step.
• New pipeline — discover → M probes → map → client checks → server → merge. CombinedResult replaces the old ValidationResult.
• Mapper daemon — persistent background process for hot model reuse (ails daemon start).
• Content-quality checks — 25 rules migrated to atom-based content queries.
• Heal command — ails heal with 4 mechanical fixers (backtick, italic, charge ordering).
• File type classification — rules target file types, skip when target surface doesn't exist.
• --strict fix — exits 1 on any finding, not just errors.

GitHub Action

• Removed invalid flags (--no-update-check, -q)
• Added parse_result.py for CombinedResult JSON parsing
• Rewritten summary.py for new schema

Release workflow

• fetch_bundled_model before build (ONNX model in wheel)
• Wheel verification step before PyPI publish
• Post-publish smoke tests: pip, uvx, npx

Breaking

• JSON schema: files/stats replaces score/level/violations
• Removed commands: update, dismiss, judge
• Removed flags: --experimental, --no-update-check, -q
• Project config dir: .reporails/ → .ails/