README
This repository employs the use of LLMs for automatic programming.
This readme is written by a human.
treesitter-index
treesitter-index is a cli tool that uses tree-sitter to generate a source code skeleton that strips most of the code implementation but retains many of the structurally useful information that can act as an index into the source code. The resulting index output is most useful to conserve tokens and improve the agentic programming experience.
Installation
In a checked out repository:
cargo install --path .
ripgrep
ripgrep (rg) is used to pre-filter files when a regex pattern is used (-e) and more than one input file or folder is provided and needs to be available on the system.
Usage
AGENTS.md
Put this instruction in your AGENTS.md
- When analyzing source files, prefer running `treesitter-index -g <glob> <files or directories...>` early to obtain a compact structural skeleton before reading the full file, and then perform targeted reads for implementation details. Line numbers are indicated in square brackets (e.g [5] means line 5, and [5-10] means lines 5 to 10). To filter for specific top-level symbols, use `-e`, for example: `treesitter-index -g "*.ts" -i -e "manager|main" src`. Add `--match-imports` to include matching imports. Don't use `*` as the glob since that circumvents the default ignore rules.
Skill file
You can use the skill in .agents/skills/treesitter-index/SKILL.md instead of the instruction as well.
Examples
Index one file:
treesitter-index src/main.rs
Index a directory recursively, only rust files:
treesitter-index -g '*.rs' src
Filter for top-level symbols (i.e. classes, class methods, functions, markdown headings):
treesitter-index -i -e 'manager|main' src
Output the treesitter syntax tree instead of a skeleton:
treesitter-index --format json src/main.rs
treesitter-index --format sexp src/main.rs
Options
| Option | Description |
|---|---|
-t, --type <TYPE> | Set the source language. Required for standard input and overrides extension-based detection. |
--format <FORMAT> | Select skeleton, json, or sexp output. Defaults to skeleton. |
-g, --glob <GLOB> | Include or exclude files. Prefix an exclusion with !; rules are applied in order. |
-e, --regexp <REGEXP> | Include matching top-level symbols. May be repeated and is available for skeleton output. |
--match-imports | Also match and include imports when using --regexp. |
-i, --ignore-case | Make regular-expression matching case-insensitive. |
-h, --help | Print command-line help. |
-V, --version | Print the version. |
Directory traversal respects standard ignore files such as .gitignore.
Supported languages are Python, JavaScript, JSX, TypeScript, TSX, Rust, Go,
Java, and Markdown (.md and .markdown).
Releasing
Release a new binary distribution on github with:
cargo release patch --no-publish --execute
Acknowledgements
This project was adapted from Maki.
License
See LICENSE and THIRD_PARTY_NOTICES.
Contributors
17 commits
See what people are saying
Similar repos
Languages
Rust
100.0%
README
This repository employs the use of LLMs for automatic programming.
This readme is written by a human.
treesitter-index
treesitter-index is a cli tool that uses tree-sitter to generate a source code skeleton that strips most of the code implementation but retains many of the structurally useful information that can act as an index into the source code. The resulting index output is most useful to conserve tokens and improve the agentic programming experience.
Installation
In a checked out repository:
cargo install --path .
ripgrep
ripgrep (rg) is used to pre-filter files when a regex pattern is used (-e) and more than one input file or folder is provided and needs to be available on the system.
Usage
AGENTS.md
Put this instruction in your AGENTS.md
- When analyzing source files, prefer running `treesitter-index -g <glob> <files or directories...>` early to obtain a compact structural skeleton before reading the full file, and then perform targeted reads for implementation details. Line numbers are indicated in square brackets (e.g [5] means line 5, and [5-10] means lines 5 to 10). To filter for specific top-level symbols, use `-e`, for example: `treesitter-index -g "*.ts" -i -e "manager|main" src`. Add `--match-imports` to include matching imports. Don't use `*` as the glob since that circumvents the default ignore rules.
Skill file
You can use the skill in .agents/skills/treesitter-index/SKILL.md instead of the instruction as well.
Examples
Index one file:
treesitter-index src/main.rs
Index a directory recursively, only rust files:
treesitter-index -g '*.rs' src
Filter for top-level symbols (i.e. classes, class methods, functions, markdown headings):
treesitter-index -i -e 'manager|main' src
Output the treesitter syntax tree instead of a skeleton:
treesitter-index --format json src/main.rs
treesitter-index --format sexp src/main.rs
Options
| Option | Description |
|---|---|
-t, --type <TYPE> | Set the source language. Required for standard input and overrides extension-based detection. |
--format <FORMAT> | Select skeleton, json, or sexp output. Defaults to skeleton. |
-g, --glob <GLOB> | Include or exclude files. Prefix an exclusion with !; rules are applied in order. |
-e, --regexp <REGEXP> | Include matching top-level symbols. May be repeated and is available for skeleton output. |
--match-imports | Also match and include imports when using --regexp. |
-i, --ignore-case | Make regular-expression matching case-insensitive. |
-h, --help | Print command-line help. |
-V, --version | Print the version. |
Directory traversal respects standard ignore files such as .gitignore.
Supported languages are Python, JavaScript, JSX, TypeScript, TSX, Rust, Go,
Java, and Markdown (.md and .markdown).
Releasing
Release a new binary distribution on github with:
cargo release patch --no-publish --execute
Acknowledgements
This project was adapted from Maki.
License
See LICENSE and THIRD_PARTY_NOTICES.
See what people are saying
Similar repos
Contributors
17 commits
Languages
Rust
100.0%