Patch mode & AI

The agent can patch your files in two ways: AST-based (syntax-aware, deterministic) or AI-based (LLM rewrites whole files). By default it uses hybrid mode.

Modes

Mode	Flag	Description
hybrid	`--patch-mode hybrid` (default)	Try AST patching first; fall back to AI if needed and configured
ast	`--patch-mode ast`	AST-only. No LLM, no API key. Best when your code layout is standard.
ai	`--patch-mode ai`	AI-first. Send entire files to an LLM for patching. Useful for complex or unusual layouts.

Example — use AI for everything:

npx @synchronized-studio/cmsassets-agent init --patch-mode ai

When to use AI mode

AST fails — e.g. non-standard formatting, mixed patterns, or edge cases the parser doesn’t handle.
Unusual structure — large files, many injection points, or custom patterns.
Extra safety — you can combine AI patching with AI verification (see below).

AI mode requires:

OPENAI_API_KEY in the environment
Optional dependency openai (install if you use AI: npm install openai).

AI verification (optional)

After patching, you can run an AI review over the changed files to catch mistakes:

npx @synchronized-studio/cmsassets-agent verify --dir .

With the right profile, the verifier uses an LLM to check that each patched file still makes sense (no broken exports, no unreachable code, correct transformer usage). Useful after apply or when you’ve run the agent in AI mode.

Summary

Default hybrid is enough for most projects.
Use --patch-mode ast for a strict, no-API, deterministic run.
Use --patch-mode ai when AST struggles or you want the model to handle tricky files.
Use verify for an extra AI check after patching.

← Commands & options · Supported frameworks →