I wrote a small CLI tool in Go called Dewdrops to help with "context stuffing" (feeding codebases to Claude/GPT).
The Problem: Copy-pasting files manually is slow, and most existing scripts don't handle .gitignore logic correctly or include binary files that waste tokens.
The Solution:
Dewdrops walks your directory tree, strictly respects your .gitignore rules (using git index logic), skips binaries/images, and outputs a formatted Markdown file.
Key Features:
Tree-First: Prints a tree structure at the top so the LLM understands architecture before reading code.
Go-based: Compiles to a single static binary (no npm/pip deps).
Air-Gap Ready: Runs entirely locally.
It’s open source (MIT). I’d love to hear your feedback on the traversal logic or formatting.
I agree that for fully autonomous agents (or IDE extensions), tool-use/search is the superior architecture. dewdrops is specifically targeting the "Chat UI" workflow, where a developer just wants to drag/drop a feature branch into Claude/GPT/Gemini to ask a quick question without setting up a local RAG pipeline. It's a "dumb tool" for a specific constraint.
it also doesn't touch or generate AGENTS.md. It just respects .gitignore and dumps the raw source.
I wrote a small CLI tool in Go called Dewdrops to help with "context stuffing" (feeding codebases to Claude/GPT).
The Problem: Copy-pasting files manually is slow, and most existing scripts don't handle .gitignore logic correctly or include binary files that waste tokens.
The Solution: Dewdrops walks your directory tree, strictly respects your .gitignore rules (using git index logic), skips binaries/images, and outputs a formatted Markdown file.
Key Features:
Tree-First: Prints a tree structure at the top so the LLM understands architecture before reading code.
Go-based: Compiles to a single static binary (no npm/pip deps).
Air-Gap Ready: Runs entirely locally.
It’s open source (MIT). I’d love to hear your feedback on the traversal logic or formatting.
1. Search is better than summaries
2. Don't automate AGENTS.md creation to someone else.
more insghts please? would be thankful
I'm not going to do that in HN comments, get on Bluesky and join the conversations there
I agree that for fully autonomous agents (or IDE extensions), tool-use/search is the superior architecture. dewdrops is specifically targeting the "Chat UI" workflow, where a developer just wants to drag/drop a feature branch into Claude/GPT/Gemini to ask a quick question without setting up a local RAG pipeline. It's a "dumb tool" for a specific constraint.
it also doesn't touch or generate AGENTS.md. It just respects .gitignore and dumps the raw source.