Drift Detection

What is rules drift?

Rules drift occurs when AI rule files become misaligned with the codebase they describe. Types of drift:

• Path drift — a referenced file path no longer exists
• Glob drift — a glob pattern matches zero files
• Command drift — a referenced command no longer works
• Coverage gap — a codebase pattern has no rule covering it
• Date drift — a rule’s validation date exceeds its cadence threshold

Why it matters

Drifted rules don’t just fail to help — they actively mislead. A path drift causes agents to look for files that don’t exist. A stale command causes failed tool invocations.

Running drift detection

# standalone drift check
bunx @lambdacurry/anvil drift ./my-repo

# drift is also included in the full audit
bunx @lambdacurry/anvil audit --target ./my-repo

What Anvil checks

1. Glob resolution — every glob pattern in .mdc files is resolved against the actual file tree
2. Path existence — referenced file paths are checked for existence
3. Validation dates — rules with Last validated headers are checked against their cadence
4. Coverage analysis — codebase patterns are checked for matching rule coverage

Interpreting drift output

Each drift issue includes:

• File and line — where in the rule file the drifted reference lives
• Type — path, glob, command, coverage, or date
• Severity — low, medium, or high based on impact
• Suggestion — what to fix or remove

Fixing drift

• Stale globs — update the glob pattern or remove the rule section
• Missing paths — update the path reference or confirm the file was intentionally removed
• Date drift — re-validate the rule and update the Last validated header
• Coverage gaps — add a new rule or confirm the pattern is out of scope