find-untested-sources

# Find Untested Sources Agent builds a cheap source-to-test pairing map before slower coverage work begins. ## When to Use | User prompt | Use | |---|---| | User asks where to add tests next | Use this skill | | User asks for files with no obvious test pair | Use this skill | | User asks to validate whether generated tests map to intended sources | Use this skill | ## When Not to Use | User prompt | Route | |---|---| | User asks for executed line or branch coverage | Use `coverage-analysis` | | User asks whether current tests are strong | Use `test-gap-analysis` or `assertion-quality` | | User asks for CRAP scores | Use `crap-score` | ## Required Inputs | Input | Required | Description | |---|---|---| | Repo root | Yes | Agent needs the analysis boundary. | | Language profile | No | Agent chooses Roslyn or tree-sitter when omitted. | | Output limit | No | Agent returns Top N candidates when brevity matters. | ## Engine Table | Engine | Agent uses it when | Strength | Agent caution | |---|---|---|---| | Roslyn script | Repo is effectively .NET-only | Namespace-aware pairing and better symbol matching | Agent avoids tree-sitter when Roslyn precision is available. | | Tree-sitter script | Repo is polyglot or one cross-language pass is needed | Broad language coverage | Agent states that heuristic precision is lower. | ## Output Table | Field | Meaning | |---|---| | `repo` or `repo_root` | Analysis root | | `summary` or `counts` | Totals for source, test, paired, and untested files | | `untested` or `untested_sources` | Prioritized candidate files | | `source_to_tests` or `tested_sources` | Known pairings | | `suggested_test_path` | Likely next test location | ## Workflow | Step | Agent action | Test | Pass | |---|---|---|---| | 1 | Agent chooses Roslyn or tree-sitter. | Review repo language mix. | Engine fits the repository. | | 2 | Agent excludes build, vendor, and generated folders. | Review discovery rules. | Worklist avoids noise. | | 3 | Agent builds the source-to-test pairing map. | Review produced output. | Source and test counts are present. | | 4 | Agent ranks untested files by declaration surface or likely impact. | Review ordering. | High-value files rise to the top. | | 5 | Agent states that results are static heuristics, not runtime coverage truth. | Review report wording. | Limitations stay explicit. | ## Verification Checklist - [ ] Agent picked the correct engine for repo language mix. - [ ] Agent excluded generated and build artifacts. - [ ] Agent returned prioritized untested candidates with suggested test paths. - [ ] Agent described the result as static pairing only. - [ ] Agent kept the output concise enough to drive next actions. ## Common Pitfalls | Pitfall | Agent fix | |---|---| | Agent presents the result as coverage proof | Agent states heuristic limits explicitly. | | Agent chooses tree-sitter for a pure C# repo without reason | Agent prefers Roslyn. | | Agent reads the whole repo manually first | Agent lets the analyzer build the initial worklist. | | Agent ignores weak pairings that deserve follow-up | Agent surfaces them as candidates for deeper review. |