- name
- sonar-coverage
- description
- Find files with low test coverage and inspect uncovered lines in a SonarQube project (project key optional when MCP integration already defines the default project)
- argument-hint
- [project-key?] [--max n] [--file key] [--pr id]
- allowed-tools
- Read, Grep, Bash(docker ps:*), Bash(podman ps:*), Bash(nerdctl ps:*), Bash(sonar:*)
# SonarQube — Coverage
Identify files with insufficient test coverage and pinpoint the exact lines that need tests.
## Usage
```
sonar-coverage # worst-covered files in the current project
sonar-coverage my-project # worst-covered files in a specific project
sonar-coverage my-project --max 50 # only files with coverage <= 50%
sonar-coverage my-project --file src/auth/login.py # line-by-line detail for one file
```
## Prerequisites
This skill requires the SonarQube MCP Server to be configured and the tools `mcp__sonarqube__search_files_by_coverage` and `mcp__sonarqube__get_file_coverage_details` to be available in your session.
**Before proceeding**, verify the tools are accessible. If they are not, try the `sonar api` CLI fallback in Step 3 before giving up — don't invent other CLI commands (e.g. `sonar mcp call` or `sonar coverage` do not exist).
**If the CLI fallback also fails (for example `sonar` not installed/authenticated, or no project key can be resolved), narrow down the cause** — check whether the `sonarqube` MCP server is enabled in this agent's configuration.
- **Not enabled / not registered** → recommend running the sonar-integrate skill.
- **Enabled but its tools are still unavailable** → configuration is correct but the server failed to start. The most common cause is that the container runtime is not running — the MCP server launches inside Docker/Podman/Nerdctl via `sonar run mcp`, so a correctly configured server still produces no tools if the daemon is stopped. Run `docker ps` yourself (falling back to `podman ps` / `nerdctl ps`) to confirm which cause applies: if it errors, the runtime is down; after the user starts it, confirm the same command succeeds before asking them to restart the agent session.
Either way, show the user:
> Unable to reach the SonarQube MCP Server, or project key not found.
>
> **Possible causes:**
> - MCP server not registered — invoke the sonar-integrate skill to configure the SonarQube MCP Server, then restart the agent session
> - Container runtime not running — the SonarQube MCP Server runs inside a container (Docker, Podman, or Nerdctl); start your container runtime, then restart the agent session
> - Credentials not configured — invoke the sonar-integrate skill
> - Project key is wrong or no default project in MCP config — pass an explicit key, or verify `sonar-project.properties` / re-run the sonar-integrate skill for this project
Then ask the user (yes/no) whether to run the sonar-integrate skill now. Briefly explain what it does: it checks the SonarQube setup on their machine — installing or updating `sonarqube-cli` and verifying authentication — and re-configures the integration for this agent, including the SonarQube MCP server and secrets-scanning hooks. If they confirm, invoke the sonar-integrate skill yourself and follow it end-to-end in this session, then ask the user to ensure a container runtime (Docker, Podman, or Nerdctl) is running and to restart the agent session so the new MCP tools become available; if they decline, stop.
## Instructions
### Step 1: Resolve the project key (only when needed)
MCP tools sometimes **do not require** `projectKey` after the sonar-integrate skill has stored the default project for this workspace. Resolve a key only when you must pass it (tool schema requires it, or the user targets another project):
- If the user provided a project key, use it.
- Otherwise look for `sonar.projectKey` in `sonar-project.properties` at the repo root.
- If still not found, **omit `projectKey`** in MCP calls and rely on the integration default.
### Step 2: Parse optional flags from the user-provided arguments
| Flag | Meaning |
| -------------- | --------------------------------------------------------------------------- |
| `--max <n>` | Only return files with coverage ≤ n% (maps to `maxCoverage`) |
| `--pr <id>` | Analyse a pull request instead of the main branch |
| `--file <key>` | Skip the file list and go straight to line-by-line detail for this file key |
### Step 3: Run the appropriate flow
#### Flow A — File list (default, no `--file`)
Call `mcp__sonarqube__search_files_by_coverage`. Include **`projectKey` only if** you resolved one in Step 1 **and** the tool requires it; otherwise omit it.
```json
{
"projectKey": "<only-if-required>",
"maxCoverage": <n>, // if --max was given
"pullRequest": "<id>", // if --pr was given
"pageSize": 20
}
```
Omit `projectKey` from the payload entirely when the default project from integration applies. Omit unused optional fields.
Present results as a table sorted by coverage ascending:
```markdown
## Coverage — `my-project`
Files with lowest coverage (worst first):
| File | Coverage |
| ------------------- | -------- |
| src/auth/login.py | 12.5% |
| src/utils/crypto.py | 23.0% |
| src/api/routes.py | 41.8% |
```
If no files are returned (all files exceed the threshold), say: *"All files meet the coverage threshold."*
Then offer to drill in:
*"Ask me to inspect any of these files for uncovered lines, or invoke the sonar-coverage skill with `--file <file-key>` (add a project key only if needed)."*
**If `mcp__sonarqube__search_files_by_coverage` is unavailable, fall back to `sonar api`.** This needs an explicit project key (no MCP default) — if none was resolved in Step 1, ask the user or invoke sonar-list-projects, then stop.
```bash
sonar api get "/api/measures/component_tree?component=<project-key>&metricKeys=coverage,line_coverage,branch_coverage&qualifiers=FIL&strategy=leaves&s=metric&metricSort=coverage&asc=true[&pullRequest=<id>]"
```
Use `metricKeys`. If a call 400s unexpectedly, add `-v` to see the actual server error. Read `coverage` from each component's `measures` array and present it in the same table format above.
#### Flow B — Line detail (`--file <key>` given, or user asks to inspect a file)
Call `mcp__sonarqube__get_file_coverage_details`:
```json
{
"key": "<file-key>",
"pullRequest": "<id>" // if --pr was given
}
```
The file key format is `<projectKey>:<path>`, e.g. `my-project:src/auth/login.py`.
If the user provides just a path, prepend the resolved project key when you have one; if the integration supplies the default project, the detail tool may accept the path or key format your MCP schema documents — follow the tool schema.
Present uncovered and partially covered lines:
```markdown
## Coverage Detail — `src/auth/login.py`
Overall coverage: **12.5%**
### Uncovered lines
Lines with no test coverage: 14, 15, 23, 45–52, 67
### Partially covered branches
| Line | Covered branches | Total branches |
| ---- | ---------------- | -------------- |
| 30 | 1 | 2 |
| 61 | 0 | 2 |
```
If the file is fully covered, say: *"All lines in this file are covered."*
**If `mcp__sonarqube__get_file_coverage_details` is unavailable, fall back to `sonar api`:**
```bash
sonar api get "/api/sources/lines?key=<file-key>[&pullRequest=<id>]"
```
Each returned line has coverage fields (covered/uncovered, branch hit counts) — derive the uncovered-lines list and partially-covered-branches table from those.
If this also fails, show the standard message above — don't guess further commands.
### Step 4: Next steps
- To write tests for uncovered lines: *"Ask me to add tests for the uncovered lines above."*
- To check for quality issues in the same file: *"Invoke the sonar-list-issues skill with `--component <file>`."*
- To check the quality gate: *"Invoke the sonar-quality-gate skill (add a project key only if you are not using the integration default)."*
GitHub에서 보기