Crate spec_drift 

spec-drift — semantic coherence analysis between a project’s specification surfaces (README, AGENTS.md, examples, CI) and its Rust code.

The library exposes the domain model, analyzers, and reporters so they can be embedded in editors or other tools. The spec-drift binary is a thin CLI wrapper over run_cli.

Re-exports

pub use config::Config;

pub use config::ConfigSource;

pub use context::ProjectContext;

pub use domain::Attribution;

pub use domain::ClaimKind;

pub use domain::CodeFact;

pub use domain::Confidence;

pub use domain::Divergence;

pub use domain::FactKind;

pub use domain::Location;

pub use domain::RuleId;

pub use domain::Severity;

pub use domain::SpecClaim;

pub use error::SpecDriftError;

Modules

analyzers

auto_fix

Auto-fix suggestions for deterministic drift rules.

baseline

--baseline support.

blame

--blame divergence attribution via git blame --porcelain.

config

spec-drift.toml configuration.

context

domain

error

llm

LLM-backed rule support.

parsers

reporters

sources

suppress

Inline-ignore filtering.

workspace

Workspace member discovery via cargo metadata.

Structs

RunConfig

Everything the orchestration layer needs to run a spec-drift analysis.

Enums

Pillar

Pillar selection mirrors the CLI --docs / --examples / --tests / --ci flags.

Functions

apply_config

Apply a Config to a divergence set: drop suppressed items and apply severity overrides.

apply_strict

Promote every non-deterministic divergence one severity level. Mirrors the --strict CLI flag — deterministic rules stay untouched because they already carry unambiguous verdicts.

normalize_locations

Render and baseline locations relative to the workspace root when possible. Analyzers work with absolute paths so suppression and blame can read files reliably; reporters should not leak machine-local temp or checkout paths.

run

Execute every analyzer in parallel and return divergences sorted deterministically by (file, line, rule) so output can be diffed between runs.

run_cli

Full end-to-end run: discover config, walk files, parse sources, select and run analyzers, apply suppression / baselines / strict, enrich with blame, and render through the chosen reporter.