Code Research

Perform deep research on "$ARGUMENTS" using ChunkHound.

Configuration

1. •Read ~/.claude/skills/chunkhound-config.json — contains embedding_args, llm_args, and db_dir.
2. •Read ~/.claude/skills/projects.json — maps project names to paths.

Steps

1. •

   Read the config and registry files listed above.

2. •

   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.

3. •

   Check if the DB exists. If {db_dir}/{name}.duckdb does not exist, index it first:

   bash

   chunkhound index {project_path} --db {db_dir}/{name}.duckdb {embedding_args}
   

4. •

   Run deep research:

   bash

   chunkhound research "$ARGUMENTS" --db {db_dir}/{name}.duckdb {embedding_args} {llm_args}
   

5. •

   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.

6. •

   If no results or the DB doesn't exist and we're not in a project, use AskUserQuestion to ask:

   • •Question: "Not inside a registered project. Which project should I research?"
   • •Options: one option per registered project (only those whose DB files exist), plus "All projects".
7. •

   If the user selects a project, run the research command against that project's DB.