check-best-practice-compliance
Check Best Practice Compliance
Run a compliance check against any artifact using the resolved effective preferences from grimoire.toml. Outputs LSP-compatible JSON (machine-readable, editor-consumable) and an HTML coverage report (human-readable).
Why This Is Best Practice
Adopted by: Linters (ESLint, Checkstyle, Rubocop), architecture validation tools (ArchUnit), and policy-as-code frameworks (OPA, Conftest) encode quality criteria once and run them repeatedly against any artifact. Google's Tricorder runs 30+ analyzers on every CL and blocks 15–20% that would have passed human review; false-positive rate kept below 5% by restricting to well-defined criteria (Sadowski et al., ICSE 2015).
Impact: Manual best-practice review is inconsistent — what gets checked depends on who reviews and when. A compliance check applies the same criteria every run and produces a structured, comparable result. Coverage metrics make progress visible: teams that track compliance coverage improve it (Hawthorne effect for quality metrics, confirmed in code review studies at Microsoft Research).
Why best: review-best-practice-fit is one-shot and subjective. check-best-practice-compliance is repeatable, explicit, and machine-readable — it runs against the user's actual stated preferences (resolved from settings), producing LSP-compatible diagnostics that any editor or CI system can consume directly.
Sources: Sadowski et al. (2015) "Tricorder: Building a Program Analysis Ecosystem Inside Google"; Ford, Parsons, Kua (2017) "Building Evolutionary Architectures"; ESLint architecture documentation
Steps
1. Resolve the effective spec (silent)
Grimoire-first (preferred): If grimoire MCP is connected, call grimoire_context — use resolved_profiles, domain_sections, and settings_sources as the authoritative resolved spec. Grimoire has already expanded profiles (5-step file search + 4-layer composition) and merged all settings layers. Do NOT re-derive what grimoire already provides.
If CLI only (no MCP): run grimoire context --json and read the same fields from the output.