Overview of the protected frontend quality thresholds and a quick-reference router that maps every failing quality check — ESLint, TypeScript, markdownlint, jscpd, rust-code-analysis metrics, dependency-cruiser, Jest/Playwright coverage and visual tests, Stryker mutation, Lighthouse, and accessibility — to its make target and specialized fixing skill. Use when you need to understand quality metrics, run comprehensive quality checks, or learn which specialized skill to use. For specific issues, use dedicated skills (complexity-management for rust-code-analysis or jscpd, architecture for dependency-cruiser, frontend-testing-workflow for coverage and mutation, frontend-performance-accessibility for Lighthouse, accessibility-audit for a11y).
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Overview of the protected frontend quality thresholds and a quick-reference router that maps every failing quality check — ESLint, TypeScript, markdownlint, jscpd, rust-code-analysis metrics, dependency-cruiser, Jest/Playwright coverage and visual tests, Stryker mutation, Lighthouse, and accessibility — to its make target and specialized fixing skill. Use when you need to understand quality metrics, run comprehensive quality checks, or learn which specialized skill to use. For specific issues, use dedicated skills (complexity-management for rust-code-analysis or jscpd, architecture for dependency-cruiser, frontend-testing-workflow for coverage and mutation, frontend-performance-accessibility for Lighthouse, accessibility-audit for a11y).
Command convention: make <make.X> means "run make with the target the
profile maps for key make.X". A null mapping means the capability is
absent in this repository — skip that check with a capability-absent note
instead of failing. Generic tooling (bun, git, gh) is invoked
directly. Never run ESLint, tsc, markdownlint, jscpd, rust-code-analysis,
dependency-cruiser, Jest, Playwright, Stryker, or Lighthouse bare on the
host: every quality tool runs through its mapped make target, which wraps
the containerized toolchain and the bun-managed dependencies. When make.a11y
is null, the accessibility lane runs through the plugin's bundled a11y check
(see accessibility-audit) — or is skipped
with a capability-absent note when capabilities.accessibility_audit is false.
Context (Input)
Need to understand the protected quality thresholds
Running comprehensive quality checks before commit
Determining which specialized skill to use for a specific issue
Quick reference for quality tool commands
When run under /fe-sdlc-finish-pr, the ci-fixer agent drives the CI loop
to green; invoked directly, follow the same steps yourself.
Task (Function)
Understand the frontend quality metrics and route to the appropriate
specialized skill for fixes.
Success Criteria: Know which skill to use for your specific quality issue.
Protected Quality Thresholds
CRITICAL — raise-only rule: the quality.* values in the project
profile are floors/ceilings over the shipped defaults. A profile may tighten
the bar (raise score floors), never relax it. Violation-count ceilings ship at
0 and may not be raised. NEVER lower any threshold in the profile or in the
tool config files (jest.config.ts, stryker.config.mjs, .jscpd.json,
config/metrics-policy.json, lighthouse/lighthouserc.*.js, eslint.config.mjs,
the dependency-cruiser config).
The ESLint gate enforces the project's no-restricted-syntax conventions —
classes-with-instance-methods (no static, no free functions) outside React
components, type-only files, and no data-testid in src/**. The metrics gate
hard-fails on the rust-code-analysis bands in config/metrics-policy.json
(cyclomatic > 10, cognitive > 15, ABC magnitude > 17, function/closure args
3, exit points > 3, function/file LLOC/PLOC/SLOC and Halstead ceilings, MI
< 20, class WMC/NPM/NPA/COA/CDA limits); those values are repository policy,
not profile keys — the raise-only rule applies to them identically: tighten if
needed, never loosen.
Jest runs two separate environments (client jsdom, server node) plus an
integration project that enforces full coverage over the source root; Stryker
mutation is sharded across a matrix and re-enforced by the merge gate
(make.test_mutation + make.merge_mutation_reports). MSI is measured only
over mutated lines, so it is a separate, stronger signal that does NOT by
itself guarantee full line coverage — the quality.coverage_* floors are an
independent bar.
Quick Reference Commands
Comprehensive checks
# Run the full local CI suite (recommended before commit)
make <make.ci>
Success: the target exits 0 and every reported score sits at or
above its quality.* floor (and every violation count at 0). Some
repositories also print a success banner — treat exit status as the contract,
the banner as confirmation.
Degrade rule (capability absent): if make.ci is null, run the
individually mapped targets in order — make.format first (mutating), then
make.lint_eslint, make.lint_tsc, make.lint_md, make.lint_dup,
make.lint_metrics, make.lint_deps, the unit/integration/E2E lanes, then the
capability-gated lanes — skipping any null entry with an explicit
capability-absent note:
SKIPPED: make.test_visual when capabilities.visual_testing is false (or make.test_visual is null).
SKIPPED: make.test_mutation + make.merge_mutation_reports when capabilities.mutation_testing is false.
SKIPPED: make.lighthouse_desktop / make.lighthouse_mobile when capabilities.lighthouse is false.
SKIPPED: make.a11y when capabilities.accessibility_audit is false.
Individual quality checks
Check
Command
Purpose
Format (preflight)
make <make.format>
Prettier + qlty fmt (mutating; runs first, alone)
Aggregate lint
make <make.lint>
Read-only lint aggregate
ESLint
make <make.lint_eslint>
Lint rules incl. no-restricted-syntax gates
TypeScript
make <make.lint_tsc>
Type checking and errors
Markdown
make <make.lint_md>
Markdown lint
Duplication (jscpd)
make <make.lint_dup>
Copy/paste DRY gate
Metrics (rca)
make <make.lint_metrics>
rust-code-analysis complexity gate
Dependencies
make <make.lint_deps>
dependency-cruiser layer/import boundaries
Accessibility
make <make.a11y>
a11y audit (bundled lane when make.a11y is null)
Testing commands
Check
Command
Purpose
Unit (client)
make <make.test_unit_client>
Component/hook tests (jsdom)
Unit (server)
make <make.test_unit_server>
Apollo mock/server tests (node)
Integration
make <make.test_integration>
Component interactions, 100% over the source root
E2E
make <make.test_e2e>
Full user flows (Playwright + Mockoon mocks)
Visual
make <make.test_visual>
Visual regression snapshots
Mutation
make <make.test_mutation> / make <make.merge_mutation_reports>
Test-quality validation (sharded + merge gate)
Lighthouse
make <make.lighthouse_desktop> / make <make.lighthouse_mobile>
Perf/a11y budget (desktop + mobile)
Routing to Specialized Skills
When quality checks fail, use the appropriate specialized skill:
Read the grouped per-target output for the specific failure, e.g. a
rust-code-analysis cognitive-complexity value reported above its
config/metrics-policy.json limit, or a Jest coverage number below the
quality.coverage_* floor.
Step 3: Use the specialized skill
Failure Pattern
Skill to Use
"cyclomatic/cognitive/ABC over limit"
complexity-management
jscpd clone found
complexity-management
dependency-cruiser violation / "must not depend on"
Repeat until the target exits 0 with every score at or above its floor
and every violation count at 0.
Constraints (Parameters)
NEVER
Lower quality thresholds in the profile or tool config files
(jest.config.ts, stryker.config.mjs, .jscpd.json,
config/metrics-policy.json, lighthouse/lighthouserc.*.js,
eslint.config.mjs, the dependency-cruiser config)
Skip failing checks to "save time"
Commit code while the make.ci target fails
Edit the dependency-cruiser config, .jscpd.json, or the metrics policy to
make violations disappear (fix the code, not the config)
Commit generated visual snapshots unless the UI change is intentional
Add suppression/ignore annotations to hide quality issues (eslint-disable,
// @ts-ignore, // @ts-nocheck, prettier-ignore,
editorconfig-checker-disable, markdownlint-disable, jscpd or
dependency-cruiser ignore directives, Stryker disable comments,
/* istanbul ignore */)
ALWAYS
Fix code to meet standards (not config to meet code)
Run make <make.ci> before creating commits
Use specialized skills for specific quality issues
Keep coverage at the quality.coverage_* floors and MSI at the
quality.mutation_msi floor over architecture.source_root
Keep cyclomatic complexity low (well under the policy limit) and components,
hooks, and helpers small
Respect the bulletproof-react module boundaries and the @/ / @auth/
path-alias convention
Treat accessibility as non-negotiable — clear the make.a11y lane
Before presenting changes, disclose any changed-file lines over 100
characters as path:line with the measured count (disclosure, not failure,
unless a gate fails)
Skip-with-note when a make.* key is null (capability absent)
Format (Output)
Generalized pass criteria, against the profile values:
