What problem does it solve? AI-generated code often ships with robotic section banners, phase headers, and docstrings that restate the signature. Manually hunting these artifacts across a branch is tedious and error-prone, while blanket comment deletion destroys load-bearing notes like workaround explanations and license headers. ## Core Features & Use Cases - Scoped scanning: Targets branch diffs by default, with working-tree, global, or explicit path/glob scope modes, filtering out generated code, vendored directories, and non-code extensions. - Parallel proposal agents: Dispatches Sonnet proposer agents in waves of six to identify high-confidence AI artifacts in comment regions only, using a per-language comment syntax reference covering Python, TypeScript, Rust, Go, C/C++, and more. - Gated apply: Consolidates all proposals into a single diff with overlap detection, presents it for explicit user approval, then applies edits surgically via the Edit tool without touching executable code or string literals. - Use Case: Before merging an AI-assisted feature branch, run decomment to strip banner dividers and restate-the-signature docstrings while preserving linter directives, license headers, and WHY comments. ## Quick Start Ask the assistant to decomment the current branch, for example by saying "strip AI comments from my branch", then review the proposed diff and approve it at the gate.
