configuration.md

Configuration reference

crie works with zero configuration — every setting below has a built-in default. Run crie init to scaffold a commented crie.config.yaml in the current directory, or point crie analyze at one explicitly with --config. Discovery order (crie/config/loader.py): an explicit --config path, then crie.config.yaml / crie.config.yml / .crie.yaml at the repository root, then built-in defaults.

Full schema

# Glob patterns (relative to the repo root) to include/exclude from analysis.
include:
  - "**/*"
exclude: []
  # - "vendor/**"
  # - "**/*.generated.*"

# Restrict analysis to specific languages. Empty = every supported language.
# Valid values: python, javascript, typescript
languages: []

# Cap how many commits git-history-based analyzers walk. Omit for full history.
max_commits: 5000

# Per-analyzer enable/disable and options. Names match `crie list-analyzers`.
analyzers:
  test_coverage:
    enabled: true
    options:
      report_path: coverage.xml
  coupling:
    options:
      min_shared_commits: 2
  git_history:
    enabled: true

scoring:
  normalization: percentile   # percentile | minmax
  weight_profile: balanced    # balanced | regression-focused | greenfield
  dimensions:
    overall_risk:
      weights:
        churn: 0.40   # overrides just this category; every other category on this
                       # dimension keeps its preset value
  severity_thresholds:
    high: 0.6   # overrides just this cutoff; moderate/critical keep their defaults

output:
  directory: crie-report
  formats: [json, markdown]
  top_n: 25

# CI gating: non-zero exit from `crie analyze` if any file's score on
# `fail_on_dimension` meets or exceeds `fail_on_score`.
fail_on_score: 0.9
fail_on_dimension: overall_risk

# Override automatic project-type detection. Omit to auto-detect. See
# docs/software-evaluation.md for the full list of profile values.
# profile: backend_service

# Select a focus preset (restricts + prioritizes analyzers). Omit to run everything.
# See docs/software-evaluation.md for what each preset changes.
# focus: debugging

Field reference

Key	Type	Default	Notes
include	list of globs	["**/*"]	Applied before exclusion; a file must match at least one include pattern.
exclude	list of globs	[]	Merged with a built-in exclude list (.git, node_modules, __pycache__, .venv/venv, dist, build, cache dirs) that's always applied, plus anything git itself ignores. Patterns match anywhere in the path — a leading **/ is optional and applied automatically if omitted.
languages	list of strings	[] (all)	One or more of python, javascript, typescript.
max_commits	int or null	null (full history)	Caps how many commits git log walks — useful on very large repositories.
analyzers.<name>.enabled	bool	true	Disable a specific analyzer.
analyzers.<name>.options	dict	{}	Analyzer-specific options — see the table below.
scoring.normalization	percentile | minmax	percentile	How raw metric values are converted to [0,1]. See scoring-model.md.
scoring.weight_profile	string	balanced	One of the named presets in crie/config/defaults.py, or a custom name you've fully defined via scoring.dimensions.
scoring.dimensions.<dimension>.weights	dict	{}	Per-category weight overrides for one dimension. Only the categories you list are overridden; everything else keeps the preset value.
scoring.severity_thresholds	dict	{}	Overrides for the Low/Moderate/High/Critical percentile cutoffs (keys moderate/high/critical; must stay strictly ascending). See scoring-model.md.
output.directory	path	crie-report	Relative to your current working directory unless absolute (not relative to the analyzed repo path).
output.formats	list	[json, markdown]	Any subset of the registered reporters (crie list-analyzers doesn't list reporters; there's no dedicated list-reporters command yet — the built-ins are json, markdown).
output.top_n	int	25	How many files/modules the Markdown "top risk" view highlights.
fail_on_score	float or null	null	If set, crie analyze exits with code 1 when any file meets or exceeds this on fail_on_dimension.
fail_on_dimension	string	overall_risk	Which of the five dimensions fail_on_score gates on.
profile	string or null	null (auto-detect)	Overrides project-type detection — see software-evaluation.md for valid values. Same as --profile.
focus	string or null	null (none)	Selects a focus preset — see software-evaluation.md for valid values. Same as --focus.

Per-analyzer options

Analyzer	Option	Default	Meaning
test_coverage	report_path	auto-detects coverage.xml / cobertura.xml at the repo root	Path (relative to repo root) to a Cobertura-format coverage report.
coupling	min_shared_commits	2	Minimum number of shared commits before two files are considered coupled.

Analyzers with no listed options (git_history, size, complexity_python, complexity_js, dependency_graph, architecture, readability, maintainability, debuggability, reliability, testing, performance_readiness) take none today.

AI configuration (for crie ask only)

Not part of crie.config.yaml — configured via environment and CLI flags instead, since it's specific to one command and shouldn't be something a checked-in config file accidentally enables for everyone who runs crie analyze:

Setting	How	Default
Enable AI synthesis	pip install "code-risk-intelligence-engine[ai]" + ANTHROPIC_API_KEY env var	disabled
Model	crie ask ... --model NAME	claude-sonnet-5
Force-disable per invocation	crie ask ... --no-ai	—

See custom-inquiries.md for full setup steps (getting a key, setting the environment variable correctly per-shell) and custom-inquiries.md#troubleshooting if crie ask isn't producing synthesized answers — it distinguishes "not configured" from "configured but failed" in its output, with the actual failure reason shown in the latter case.

CLI overrides

Every CLI flag on crie analyze overrides the corresponding config value for that run only — the config file itself is never modified. See cli-reference.md.