https://github.com/thecodedrift/ast-testing
An experiment on the capabilities of AST to steer agentic systems
https://github.com/thecodedrift/ast-testing
Last synced: 9 months ago
JSON representation
An experiment on the capabilities of AST to steer agentic systems
- Host: GitHub
- URL: https://github.com/thecodedrift/ast-testing
- Owner: thecodedrift
- License: mit
- Created: 2025-09-05T01:00:37.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2025-09-12T04:19:24.000Z (9 months ago)
- Last Synced: 2025-09-12T06:22:48.847Z (9 months ago)
- Language: TypeScript
- Size: 6.84 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# AST As an Agentic Code Tool
> Can AST tools's make the world of Agentic Code better?
# About This Repo
This repo simulates a simple monorepo with multiple languages as a testbed for AST tools and agentic code capabilities.
Rule JSON Schema is @ https://raw.githubusercontent.com/ast-grep/ast-grep/main/schemas/rule.json
# AST Capabilities
- Per-editor configured, often globally installed
- vscode extensions (gives hinting info)
- Polyglot solutions exist (ast-grep)
- Cannot cross file boundaries - more limited than we thought
- Rules are per language - the pattern syntax is too
- Will mean that you cannot template this easily
- pattern: `string` is more portable, but is highly language syntax dependent
## Open Questions
- none
# Agentic Capabilities
- Terrible at writing rules. Absolutely terrible.
## Copilot / VSCode
- Can see local lint errors if enabled via an extension
- Can fix errors if it can see the errors
- Does not need `instructions.md` to trigger linting as long as the language server is enabled
## Claude Code
- Can be made more deterministic with linting instructions in the claude config files
## Cursor Code
TBD
## Codex
TBD