Deep code research using ChunkHound — investigates architecture, patterns, and relationships across the codebase
Perform deep research on "$ARGUMENTS" using ChunkHound.
~/.claude/skills/chunkhound-config.json — contains embedding_args, llm_args, and db_dir.~/.claude/skills/projects.json — maps project names to paths.Read the config and registry files listed above.
Resolve the current project from your working directory by matching against the project paths in the registry. If the cwd is inside a project path, that's the project.
Check if the DB exists. If {db_dir}/{name}.duckdb does not exist, index it first:
chunkhound index {project_path} --db {db_dir}/{name}.duckdb {embedding_args}
Run deep research:
chunkhound research "$ARGUMENTS" --db {db_dir}/{name}.duckdb {embedding_args} {llm_args}
Present the findings to the user. The research output is a comprehensive markdown analysis — summarize the key findings and include the most relevant details. Read any files referenced in the output if more context would help answer the user's question.
If no results or the DB doesn't exist and we're not in a project, use AskUserQuestion to ask:
If the user selects a project, run the research command against that project's DB.