Devsentinel

Coding agents often break complex projects because they lack context on how a single file edit affects the rest of the codebase. DevSentinel provides the missing safety layer for autonomous coding by performing static analysis and visual audits before changes are committed. It identifies every file that imports your target code, detects breaking API changes before they reach the disk, and runs visual regression tests to ensure UI edits do not break layouts.

What it does

• Impact analysis maps the blast radius of any edit to see which consumers and tests are affected.
• Contract validation identifies breaking changes in exported functions, classes, or types before you write to disk.
• Visual regression captures before-and-after screenshots to detect pixel shifts, horizontal overflow, and text overlaps.
• Safe checkpoints creates full-state backups of the workspace to allow instant rollbacks of failed experiments or refactors.

How it works

1. Scan the radius by calling analyze_blast_radius on a file or specific symbol to identify all direct and transitive dependencies.
2. Verify the contract using detect_breaking_changes with your proposed code to catch signature mismatches before they cause CI failures.
3. Audit the UI by running capture_ui_state to establish a baseline, then diff_ui_visuals after the edit to confirm layout integrity.
4. Manage state through create_checkpoint and rollback_checkpoint to safely test risky refactors without touching git history.

Frameworks & tools

This skill interfaces with an MCP server (stdio/JSON-RPC). It supports TypeScript, JavaScript, and Python. Visual audits require Playwright and Chromium.

Why this beats prompting it yourself

Standard LLM prompts cannot accurately track barrel file re-exports five hops deep or perform pixel-by-pixel comparisons of rendered DOM elements. This tool provides deterministic data from the actual filesystem and browser, replacing agent guesswork with verifiable facts.

Use cases

• Refactoring shared utility functions used across a large monorepo.
• Upgrading dependencies or running codemods with a guaranteed safety net.
• Verifying that CSS changes do not cause unintended layout shifts on deep-linked pages.
• Protecting against silent failures in Python keyword arguments during symbol renames.

Known limitations

Supports TS, JS, and limited Python regex scanning. Does not follow tsconfig extends chains or resolve computed dynamic imports. Checkpoints are limited to workspaces under 250MB.