| name | detect-monorepo-projects |
| description | Identify the SonarQube Cloud projects to create for a repository by finding each build root — the top-most build manifest that aggregates the modules building together (Maven parent with sub-modules, Gradle settings with include(...), npm/pnpm/Yarn workspace root, Nx workspace, uv workspace, or any standalone build manifest). Requires a SonarQube Cloud organization key; emits one JSON entry per build root with projectKey prefixed by that key, including for single-project repos. Use when the user says /detect-monorepo-projects, asks to detect a monorepo, or asks to map a repository to SonarQube Cloud projects. |
| user-invocable | true |
| argument-hint | <organization-key> [path to target repo] — path defaults to the current working directory |
| allowed-tools | Read, Glob, Grep, Bash, Write |
Detect Monorepo Projects
Goal: Identify each build root in the target repository — the top-most build manifest in a directory tree, i.e. a manifest that is not itself referenced as a sub-module / workspace member of another manifest higher up — and emit one SonarQube Cloud project entry per build root.
A SonarQube project corresponds to one buildable unit: a Maven multi-module parent (with its <modules>), a Gradle root build (with its included sub-projects), an npm/pnpm/Yarn workspace root (with its members), an Nx workspace, a uv workspace, or a standalone single-project package. Sub-modules of any of these are part of their build root and produce no separate entries.
A repository can have multiple build roots, especially polyglot ones (e.g. a Maven multi-module at the repo root and a standalone Python lambda inside one of its sub-module directories that is not declared in any workspace).
The output is always an array of projectKey / projectName objects, even when only one build root is found:
[{"projectKey":"acme_backend","projectName":"Backend"}]
This skill produces exactly that array.
Step 1: Resolve the organization key and target repository
Parse $ARGUMENTS. The first token is the SonarQube Cloud organization key; the second token (optional) is the path to the repository to scan.
Organization key:
- If a first token was supplied, use it as the org key.
- If no first token was supplied, ask the user:
"What's the SonarQube Cloud organization key for these projects? It will be prefixed onto every generated projectKey so they're globally unique on import (e.g. acme)."
Wait for the answer before scanning.
- Validate the org key matches
^[a-z0-9][a-z0-9_-]*$. If it doesn't, explain the required format and re-prompt.
Repository path:
- If a path token was supplied, use it as the repo root.
- If no path token was supplied, use the current working directory as the repo root.
Confirm the repo root directory exists and is readable before continuing. From here on, all detection is relative to the repo root.
Step 2: Find every build root
Run every rule below. Each rule looks for the top-most build manifest of one build system and records the directory containing it. A repository can have multiple build roots in different build systems — a Maven multi-module at the repo root and a standalone Python lambda inside one of its sub-module directories, for example — so always complete all of Step 2a–2h before moving on.
Sub-modules never become their own build root. When a manifest declares its members (<modules>, include(...), "workspaces" globs, [tool.uv.workspace].members), resolve those entries to directories and add each resolved directory to a shared covered set. A directory in the covered set belongs to its parent build root and is never emitted as its own entry, regardless of which other rules might match it.
Each build root entry records:
- path — the build-root directory, relative to the repo root.
- build system — Gradle, Maven, npm, pnpm, Yarn, Nx, uv, Poetry, setuptools.
- name — derived from the manifest at that path (see each rule's naming bullet).
2a. Gradle
Start by enumerating every settings file in the repo — at any depth. Nested or sibling Gradle builds are common (e.g. a libs/<library>/settings.gradle living next to a top-level settings.gradle is two independent Gradle builds, not one). Run the find below and capture every match — do not stop after the first hit, do not assume there's only one Gradle build.
find . \( -name .git -o -name node_modules -o -name build -o -name .gradle -o -name target -o -name dist \) -prune \
-o \( -name settings.gradle -o -name settings.gradle.kts \) -print
Every directory containing one of these files is a candidate build root — treat the find results as a flat list of candidates. For each candidate, parse its settings file:
- Resolve every
include(...) / include '...' declaration to a directory. Default mapping: :libs:core → libs/core, 'frontend' → frontend.
- Honor
project(':<name>').projectDir = file('<path>') overrides — they re-route an include to a different directory. Example: include 'auth-service' followed by project(':auth-service').projectDir = file('common/auth-service') resolves to common/auth-service, not auth-service.
- Resolve
includeBuild(...) entries the same way — composite builds belong to the same logical build root, so add their target directories to the covered set too.
Add every resolved directory (paths relative to the candidate's settings file) to the covered set.
After all candidates are processed, drop any candidate whose own directory is in the covered set of another candidate — that one is a sub-build included by an outer settings file. The remaining candidates are Gradle build roots.
Name: rootProject.name = "..." (or rootProject.name '...') from the candidate's settings file; fall back to the candidate directory name.
Worked example. A repo has settings.gradle at its root (including a handful of sub-modules under core/ and various service directories) AND a parallel directory libs/ with three siblings — auth/, data/, messaging/ — each containing its own settings.gradle (with rootProject.name = '<sibling>') and build.gradle. None of the three are referenced in the root's include(...) or includeBuild(...) list. Result: four Gradle build roots in the output — the repo root plus one entry per libs/* sibling. Failing to enumerate the nested settings files is the most common miss for this rule.
2b. Maven
Find every pom.xml at any depth (skip target, node_modules, .git, build, dist).
- Each directory containing a
pom.xml is a candidate.
- For each candidate, if its
pom.xml declares <modules> with <module>…</module> entries, resolve each entry to a directory (relative to the pom's own directory) and add it to the covered set. Recurse: if a child pom also declares <modules>, continue adding its sub-paths to the covered set.
- After every candidate is processed, drop any candidate whose own directory is in the covered set (it's a Maven sub-module of another candidate, possibly transitively).
- Name: read the candidate's
<name> (or <artifactId>) from the root <project> element; fall back to the directory name.
find . \( -name .git -o -name node_modules -o -name target -o -name build -o -name dist \) -prune \
-o -name pom.xml -print
2c. npm workspaces
- Find every
package.json (skip node_modules, .git, dist, build) whose JSON has a top-level "workspaces" array (e.g. ["packages/*", "apps/*"]).
- The directory containing that
package.json is a build root. Resolve every workspace glob to a directory containing a package.json and add each to the covered set. Workspace members are not separate SonarQube projects.
- Name:
name of the build-root package.json; strip any @scope/ prefix; fall back to the directory name.
2d. pnpm workspaces
- Find every
pnpm-workspace.yaml at any depth. The directory containing it is a build root.
- Resolve every glob in its top-level
packages: list and add the resolved directories to the covered set.
- Name:
name of the sibling package.json (next to the pnpm-workspace.yaml); fall back to the directory name.
2e. Yarn workspaces
- Find every
package.json whose "workspaces" is either an array or an object with a packages array. A sibling yarn.lock disambiguates Yarn from npm, but the resolution is identical to 2c.
- Name: same as 2c.
A directory matched by multiple workspace rules (e.g. package.json "workspaces" and a pnpm-workspace.yaml) is one build root, not several — de-duplicate by path in Step 3.
2f. Nx workspace
- Find every
nx.json at any depth. The directory containing it is a build root.
- Mark every directory referenced in the workspace (via
project.json files, the nx.json projects map, or a legacy workspace.json) as part of the covered set. Nx commonly layers on top of an npm/pnpm/Yarn workspace, so the covered set typically overlaps with 2c–2e — that's fine, de-duplication in Step 3 collapses them.
- Name: top-level identifier in
nx.json if present, otherwise the name of the sibling package.json, otherwise the directory name.
2g. Standalone npm / Yarn package
After 2c, 2d, 2e, and 2f have run and the covered set is finalized, also pick up any package.json not in the covered set as its own build root — this catches single-project npm/Yarn repos and sibling JS apps that don't declare any workspace. This rule mirrors Python's "standalone package" branch in 2h: workspace rules emit one entry for the workspace root, but a package.json that isn't in any workspace at all would fall through entirely without this rule.
- Guard against config-only files. Require substantive source next to the
package.json: at least one src/, lib/, app/, or pages/ sibling directory, or a top-level index.{js,ts,jsx,tsx,mjs} file. A package.json whose only content is a dependencies block — no source — is configuration (e.g. holding a CDK version for a Python project), not a project. Skip it.
- Name:
name field of the package.json; strip any @scope/ prefix; fall back to the directory name.
Worked example. A repo has a Java Gradle multi-project at the root and a standalone JavaScript app at tools/admin-ui/ (its own package.json + yarn.lock + src/, no workspaces field, not referenced anywhere else). 2a captures the Gradle root; 2c–2f find nothing (no workspaces declared); 2g picks up the admin UI as its own build root.
2h. Python (uv workspace, Poetry, setuptools, PEP 621)
Python has no single universal monorepo manifest. Handle two shapes:
- uv workspace: a
pyproject.toml at any depth with a [tool.uv.workspace] table and a members = [...] list of globs. The directory containing it is a build root. Resolve every glob to a directory containing a pyproject.toml; add those directories to the covered set.
- Standalone Python package: any directory containing
pyproject.toml, setup.py, or setup.cfg that is not in the covered set (and is not itself a uv workspace root) is its own build root. This includes directories nested inside another build system's sub-module — e.g. a Python lambda at services/foo/foo-infra/pyproject.toml is its own build root even when its sibling services/foo/foo-app is covered by a Gradle root above.
Skip vendor / build dirs while scanning: .venv, venv, node_modules, build, dist, .tox, target, .git.
find . \( -name .venv -o -name venv -o -name node_modules -o -name build -o -name dist -o -name .tox -o -name target -o -name .git \) -prune \
-o -name pyproject.toml -print -o -name setup.py -print -o -name setup.cfg -print
- Name:
[project].name (or [tool.poetry].name) in pyproject.toml; name in setup.cfg's [metadata]; name= argument in setup.py; fall back to the directory name. For a uv workspace root, prefer [project].name of the workspace-root pyproject.toml when present, else the directory name.
A single root pyproject.toml with no workspace table and no other package directories is one build root, not "no result" — emit it as a single-project repo.
Step 3: Collect build roots and de-duplicate
Merge the results of all rules into a single list keyed by source path relative to the repo root. When the same directory was reported by more than one rule (e.g. an Nx workspace layered on top of pnpm, or a JS workspace root that also has its own pyproject.toml), keep one entry and prefer the most specific declared name.
Continue to Step 4 unconditionally — the JSON is emitted even when only one build root was found. The only "no output" branch is when zero build roots were detected; see Step 6.
Step 4: Derive projectKey and projectName
For each build root, derive the two fields deterministically:
projectName — human-readable label. Take the declared name (root package.json name, Maven <name>/<artifactId>, Gradle rootProject.name, Python [project].name); otherwise the last segment of the source path. Strip any npm scope (@org/frontend → frontend) and title-case it (frontend → Frontend, my-api → My Api).projectKey — derive a slug from the base name: lowercase it, replace every run of non-alphanumeric characters with a single _, trim any leading/trailing _, then prepend <org-key>_. Examples for org key acme: frontend → acme_frontend, @org/web-app → acme_web_app, project-bindings → acme_project_bindings.
Uniqueness: projectKey values must be unique within the array. If two build roots produce the same slug, qualify the colliding slugs with their parent path segment before prepending the org key (e.g. services/a/api and services/b/api → acme_a_api and acme_b_api).
Step 5: Validate, write, and echo the JSON
- Build the array — an array of objects each with only
projectKey and projectName, sorted alphabetically by projectKey (case-insensitive ascending), regardless of detection order.
- Validate that the result is parseable JSON before writing (e.g. pipe through a JSON parser).
- Write it to
sonar-monorepo-projects.json at the repo root.
- If that file already exists, show the current contents and the proposed contents, and ask the user to confirm before overwriting.
- Echo the written JSON in the chat, plus a summary that states the organization key used and breaks down the count per build system (e.g. "Organization: acme | Gradle: 1, Maven: 1, npm: 0, Python: 2 → 4 build roots"). The org key line lets the user confirm it before acting on the file; the per-system breakdown makes it obvious if a build system was scanned but produced nothing.
- End with a "checked for" recap so the user can spot misses: Gradle (
settings.gradle[.kts]), Maven (pom.xml), npm/Yarn (package.json "workspaces" plus standalone package.json with source), pnpm (pnpm-workspace.yaml), Nx (nx.json), Python ([tool.uv.workspace] plus standalone pyproject.toml / setup.py / setup.cfg). Build systems not scanned (Bazel, Pants, Buck, Cargo workspaces, Go modules with multiple go.mod, .NET solutions, etc.) are out of scope today — invite the user to share the layout if one of those is in play.
printf '%s' "$JSON" | python3 -m json.tool > /dev/null && echo "valid JSON"
Step 6: Zero-build-roots fallback
When no build root was detected by any rule, do not write any file. Report plainly, for example:
No build manifests detected in <repo root>. Checked for: Gradle (settings.gradle[.kts]), Maven (pom.xml), npm/Yarn (package.json "workspaces" plus standalone package.json with source), pnpm (pnpm-workspace.yaml), Nx (nx.json), Python ([tool.uv.workspace] plus standalone pyproject.toml / setup.py / setup.cfg). This repo appears to be a documentation-only, scripts-only, or unsupported-build-system repository.
This detection is best-effort — invite the user to share the layout if they believe a build root exists that the standard patterns missed (e.g. Bazel WORKSPACE/MODULE.bazel, Cargo workspace Cargo.toml, .NET .sln, Go modules).
