openrewrite-analyze-recipes

Analyze OpenRewrite Recipes

Inspect the repository before choosing commands. Follow the Maven CLI skill whenever invoking Maven.

Discover the invocation

• Read the rewrite-maven-plugin configuration and record <activeRecipes> and <activeStyles>.
• Prefer repository-provided just recipes when they configure profiles, dependencies, or toolchains.
• Prefix commands with mise exec -- when the repository uses mise, and trust the checked-in config first when required.
• When calling Maven directly, activate any profile that contains the rewrite plugin.
• Never use Maven offline mode for this workflow.

Capture the dry run

Create .llm/ when needed and capture the complete output rather than only the terminal tail. Use the repository command or the equivalent Maven goal:

mvn rewrite:dryRun 2>&1 | tee .llm/rewrite-dryrun.log

Treat OpenRewrite's "Applying recipes would make changes" result as a successful violation report. If the run fails with a RocksdbMavenPomCache serialization error, rerun with -Drewrite.pomCacheEnabled=false; clearing the cache alone does not prevent recurrence.

Copy target/rewrite/rewrite.patch to .llm/rewrite-dryrun.patch when the patch exists.

Rank violations

Parse the recipe tree associated with every changed file. Count the most specific named leaf recipe in each branch. Prefer a named wrapper over a raw parameterized child recipe because a child displayed with inline options cannot be activated by name alone.

Save a two-column count and fully qualified recipe report in .llm/rewrite-violations-per-rule.txt. Report:

• Total changed files
• Patch line count
• Total violations
• Unique triggered recipes
• Every recipe ranked by count

Mark composite recipes only when the log provides enough tree context to distinguish them reliably.

Optionally run one recipe

Present the highest-ranked leaf recipes and ask the user which recipe, if any, to apply. Before applying it, require a clean working tree; do not stash or discard changes.