AgentSkillsCN

gitingest

使用 GitIngest 将任意 Git 仓库转换为专为 LLM 优化的文本文件。当用户希望导入仓库、生成代码库的文本摘要、为 LLM 分析做好准备，或需要将 GitHub URL 转换为可读的文本文件时，可使用此技能。

SKILL.md
--- frontmatter
name: gitingest
description: Convert any Git repository into a text file optimized for LLM consumption using GitIngest. Use when the user wants to ingest a repo, create a text digest of a codebase, prepare a repository for LLM analysis, or needs to convert a GitHub URL to a readable text file.

GitIngest

Convert any Git repository into a prompt-friendly text file for LLM consumption. GitIngest extracts the structure and contents of a repository into a single text file that can be easily processed by language models.

When to Use

  • •User wants to analyze an entire codebase with an LLM
  • •User needs a text representation of a repository
  • •User mentions "ingest", "digest", or converting a repo to text
  • •User wants to prepare code for LLM context

Required Package

This skill requires the gitingest Python package.

Workflow

1. Check if GitIngest is Available

First, check if gitingest is already installed and accessible:

bash
which gitingest || echo "NOT_FOUND"

Also check common venv locations:

bash
ls ~/.venvs/gitingest/bin/gitingest 2>/dev/null || echo "NOT_IN_VENVS"

2. If Not Installed, Ask User

If gitingest is not found, ask the user how they'd like to install it:

Ask: "The gitingest package is required but not installed. How would you like to install it?"

Options:

  • •Specify path - User provides path to existing venv or installation
  • •Create a venv for me - Convenience option: look up the venv-manager skill and follow its conventions

If user chooses the convenience option:

  1. •Look for a venv-manager skill by name/description
  2. •If found, follow its conventions to create the venv and install the package
  3. •If NOT found, inform the user: "The venv-manager skill isn't available. Would you like to specify a path manually, or should I create a venv at ~/.venvs/gitingest/?"

3. Identify the Target

Once gitingest is available, determine what the user wants to ingest:

  • •Local directory: A path on the filesystem
  • •GitHub URL: A repository URL like https://github.com/owner/repo
  • •Current directory: If unspecified, confirm with user

4. Run GitIngest

Run gitingest using the path determined in steps 1-2. Examples use <gitingest> as placeholder for the actual path:

For a local directory:

bash
<gitingest> /path/to/repository -o output.txt

For a GitHub repository:

bash
<gitingest> https://github.com/owner/repo -o output.txt

For the current directory:

bash
<gitingest> . -o output.txt

5. Common Options

OptionDescription
-o <file>Output to specified file (use - for stdout)
-t <token>GitHub token for private repos
--include-gitignoredInclude files normally ignored by .gitignore
--include-submodulesProcess git submodules

Private repositories: If the user needs to ingest a private repo, they must provide a GitHub token:

bash
<gitingest> https://github.com/owner/private-repo -t <GITHUB_TOKEN> -o output.txt

Or set the environment variable:

bash
export GITHUB_TOKEN=<token>
<gitingest> https://github.com/owner/private-repo -o output.txt

6. Output

GitIngest produces a text file containing:

  • •Repository structure (directory tree)
  • •File contents with clear delimiters
  • •Token count estimate (useful for LLM context limits)

After running, confirm success and report:

  • •Output file location
  • •Approximate size/token count if available
  • •Any warnings or skipped files

Tips

  • •Output files can be large for big repositories - warn user about potential size
  • •Token counts help users understand if the output will fit in LLM context windows
  • •For very large repos, consider suggesting the user focus on specific subdirectories
  • •The output is optimized for LLM consumption but is also human-readable

Reference