Detect intent drift in AI-assisted development.
A dependency-free Python library - with a thin Node CLI - that compares the original goal against the current plan and execution activity, returning an explainable alignment report.
pip install intent-driftSee a real run
One command compares your original goal against what you are actually doing, then returns a score, the evidence behind it, and what to do next.
Intent Drift vs flying blind
Every agent run produces a measurable alignment report. The alternative is discovering the drift after it ships.
Without intent drift
Drift is invisible until it surfaces as a bad PR.
With Intent Drift
Every run answers one question: are we still on target?
| Dimension | With Intent Drift | Flying blind |
|---|---|---|
| Drift detection | ✓9 weighted providers score every run | Hope the agent stays on track |
| Explainability | ✓Named failed providers + recommendation | “Something went wrong”, if you notice at all |
| Scope creep | ✓Caught early by scope_provider | Discovered after it ships |
| Architecture changes | ✓Flagged by architecture_provider | Surprise refactor in the PR diff |
| Dependency drift | ✓Monitored by dependency_provider | Unexpected npm install in the commit log |
| Audit trail | ✓Timestamped alignment history per run | No record of intent vs execution |
| Cost | ✓~1,500 tokens per run (~$0.002) | Cost of unwinding weeks of wrong work |