Skip to content

CONFIGURATION.md

Latest commit

 

History

History
335 lines (266 loc) · 7.95 KB

CONFIGURATION.md

File metadata and controls

335 lines (266 loc) · 7.95 KB
Raw

Configuration Guide

AI-SLOP Detector can be customized using .slopconfig.yaml in your project root.

Basic Configuration

Create .slopconfig.yaml:

# Metric weights
weights:
  ldr: 0.40        # Logic Density Ratio (40%)
  inflation: 0.30  # Jargon/Buzzword Inflation (30%)
  ddc: 0.20        # Dependency Check (20%)
  purity: 0.10     # Critical pattern purity (10%)

# Thresholds
thresholds:
  ldr:
    critical: 0.30    # Below this = critical
    warning: 0.60     # Below this = warning

  inflation:
    critical: 1.0     # Above this = critical
    warning: 0.5      # Above this = warning

  ddc:
    critical: 0.50    # Below this = critical (50% unused)
    warning: 0.70     # Below this = warning (30% unused)

# Pattern control
patterns:
  disabled:
    - todo_comment      # Ignore TODO comments
    - pass_placeholder  # Allow pass statements

# File exclusions
ignore:
  - "tests/"
  - "**/*_test.py"
  - "venv/"
  - ".venv/"

advanced:
  # Cross-file structure ("coherence") is measured exactly up to this many files.
  exact_topology_ceiling: 300
  # Past that ceiling, fall back to deterministic sampling. Set "exact" to force full analysis.
  topology_mode_above_ceiling: deterministic_approximate
  # Reuse analysis results for unchanged files between runs, so re-scans are faster.
  analysis_cache_enabled: true
  # How many recent commits to read when judging how often a file changes (churn).
  churn_commit_window: 200
  # Optional coverage file; when present, low-coverage files rank higher as hotspots.
  coverage_data_file: ".coverage"
  # Max number of priority "hotspot" files listed in the report.
  hotspot_limit: 10

architecture:
  enabled: false
  preset: none
  layers: []

architecture is opt-in. If you do nothing, boundary-violations reports import cycles only.

Config Validation (v3.7.2)

.slopconfig.yaml is validated by Pydantic v2 schemas before merging into the default config. Invalid values raise ValueError with the exact field path — before they can reach the GQG formula (the weighted score combiner) or the local self-calibration (a repository-scoped review-sensitivity aid based on your local history, not external validation).

Validated Sections

weights: block — each key must be float in [0.0, 1.0]:

# Raises ValueError at load time:
weights:
  ldr: 2.5        # Error: ldr must be <= 1.0
  inflation: -0.1 # Error: inflation must be >= 0.0

patterns.god_function: block:

patterns:
  god_function:
    complexity_threshold: 10  # must be int >= 1
    lines_threshold: 50       # must be int >= 1
    domain_overrides:
      - function_pattern: "train_*"   # must be string
        complexity_threshold: 30      # must be int >= 1
        lines_threshold: 200          # must be int >= 1

Error Format

All sections are validated before raising — you see every problem at once:

ValueError: .slopconfig.yaml validation failed:
  - weights: 1 validation error for _WeightsSchema
    ldr
      Input should be less than or equal to 1 [type=less_than_equal, input_value=2.5]
  - patterns.god_function: 1 validation error for _GodFunctionSchema
    domain_overrides.0.function_pattern
      Input should be a valid string [type=string_type, input_value=123]

Unknown top-level keys are silently ignored for forward compatibility.

Full specification: docs/SCHEMA_VALIDATION.md

Integration Test Configuration (v2.6.2)

Configure integration test detection:

evidence:
  tests:
    # Integration test detection
    integration:
      enabled: true
      # Directory path patterns
      dir_patterns:
        - "tests/integration/**"
        - "integration_tests/**"
        - "tests/e2e/**"
        - "e2e/**"
      # File name patterns
      file_patterns:
        - "test_integration_*.py"
        - "*_integration_test.py"
        - "it_*.py"
      # Pytest markers to detect
      pytest_markers:
        - "integration"
        - "e2e"
        - "slow"
      # Runtime signals detection
      runtime_signals:
        enabled: true
        keywords:
          - "testcontainers"
          - "docker-compose"
          - "TestClient"

# Quality claims validation (v2.6.2)
claims:
  production_ready:
    require_integration_tests: true
    require_unit_tests: true
    require_error_handling: true
    require_logging: true

  enterprise_grade:
    require_integration_tests: true
    require_monitoring: true
    require_security: true

Domain-Aware Initialization (v3.2.0)

Instead of writing .slopconfig.yaml by hand, run:

slop-detector --init
slop-detector --init --domain general
slop-detector --init --adaptive-init --init-preview
slop-detector --init --adaptive-init --apply-init-suggestions

This generates a fully-documented config tuned to your domain profile.

The adaptive layer is conservative by design:

  • preview mode prints suggestions only
  • apply mode is explicit opt-in
  • existing handwritten sections are preserved during merge
  • architecture stays disabled unless repository evidence is strong

Built-in Domain Profiles

Domain --domain key Typical use case
General general General-purpose Python libraries and scripts
Scientific / Numerical scientific/numerical NumPy, SciPy, scientific computing
Web / API web/api FastAPI, Flask, API services
Library / SDK library/sdk Shared packages and client libraries
CLI Tool cli/tool Command-line applications
Bio bio Bioinformatics and lab pipelines
Finance finance Quant and financial analysis tooling

Each profile comes with pre-configured capability_vector (weight anchors) and domain_overrides (pattern severity adjustments). The capability_vector is used by the self-calibration engine as the initial anchor for grid search.

Example generated config for a general project:

# Generated by slop-detector --init --domain general
# Auto-added to .gitignore to protect your weakness map

domain: general

weights:
  ldr: 0.40
  inflation: 0.30
  ddc: 0.20
  purity: 0.10

capability_vector:
  ldr: 0.40
  inflation: 0.30
  ddc: 0.20
  purity: 0.10

thresholds:
  ldr:
    critical: 0.30
    warning: 0.45
  inflation:
    critical: 2.0
    warning: 1.0
  ddc:
    critical: 0.30
    warning: 0.50

patterns:
  disabled:
    - todo_comment
architecture:
  enabled: true
  preset: layered
  layers: []

The built-in layered preset is intentionally narrow and evidence-heavy:

  • api -> domain is allowed
  • domain -> data is blocked
  • domain -> api is blocked
  • domain -> service is blocked

Findings are emitted as layer_boundary_violation issues inside the existing cleanup contract.

Configuration Priority

  1. Command-line arguments (highest)
  2. .slopconfig.yaml in current directory
  3. --config specified file
  4. Default values (lowest)

Example Configurations

Strict Mode (Production)

weights:
  ldr: 0.50
  inflation: 0.30
  ddc: 0.20

thresholds:
  ldr:
    critical: 0.40
    warning: 0.70
  inflation:
    critical: 0.8
    warning: 0.4
  ddc:
    critical: 0.60
    warning: 0.80

patterns:
  disabled: []  # Enable all patterns

Lenient Mode (Development)

thresholds:
  ldr:
    critical: 0.20
    warning: 0.50
  inflation:
    critical: 1.5
    warning: 0.8

patterns:
  disabled:
    - todo_comment
    - fixme_comment
    - pass_placeholder

Dependency Hygiene + Architecture Review

ignore:
  - "tests/**"
  - "node_modules/**"

advanced:
  coverage_data_file: ".coverage"
  churn_commit_window: 200

architecture:
  enabled: true
  preset: layered
  layers: []

This enables:

  • manifest-level unused-deps review for pyproject.toml and package.json
  • undeclared import detection
  • opt-in layered boundary review on top of default cycle detection

See Also