Detect Static Dependencies

Detect Static Dependencies

Scan a C# codebase for calls to hard-to-test static APIs and produce a ranked report showing which statics appear most frequently, which files are most affected, and which abstractions already exist in the .NET ecosystem to replace them.

When to Use

• Auditing a project's testability before adding unit tests
• Understanding the scope of static coupling in a legacy codebase
• Prioritizing which statics to wrap first (highest-frequency wins)
• Creating a migration plan for incremental testability improvements

Response Guidelines

• Scale the response to the user's request. A question about a specific category (e.g., "find time statics") should focus on that category with file locations and counts, not produce a full report across all categories.
• When the user provides a specific file or directory path, scan only that scope — do not expand to the entire solution unless asked.
• The full structured report format in Step 4 is for comprehensive audit requests. For focused questions, return only the relevant subset (e.g., category summary + affected files for the requested category).

Execution Contract

• A relative path named in the prompt is enough to start. Discover it with the available file-listing tools and scan it immediately; do not ask the user to provide or re-upload files before both discovery and a content search fail.
• Start with a recursive, line-numbered content search over eligible .cs files. Do not search only for the static keyword: ambient calls inside LINQ expressions, lambdas, callbacks, and interpolated strings usually have no static modifier.
• If a file-reading tool fails on a path that listing or search proved exists, classify the failure before retrying. Fall back to another available mechanism such as rg -n, grep, or a shell file reader only for confirmed tool availability, transport, or path-normalization failures and only after verifying the canonical path remains inside the workspace. Stop on content-exclusion, permission/policy, workspace-boundary, or unknown failures. Search output can seed the occurrence ledger; open only the surrounding code needed to verify receiver provenance.
• Never stop after loading this skill or announcing a scan plan. Return the completed audit in the same response. If every fallback genuinely fails, report the verified partial findings and the exact limitation; do not invent findings or replace the audit with a request to rerun.

When Not to Use

• The user wants wrappers generated (hand off to generate-testability-wrappers)
• The user wants mechanical migration done (hand off to migrate-static-to-wrapper)
• The statics are already behind interfaces or TimeProvider
• The code is not C# / .NET

Inputs

Workflow

Step 1: Determine scan scope

Resolve the target to a set of .cs files:

• Treat a prompt-named workspace-relative path as the target; locate it rather than asking the user for an absolute path.
• If omitted, scan every eligible .cs file under the current workspace; do not pick one project and silently omit its siblings.
• If a .cs file, scan that single file.
• If a directory, scan all .cs files recursively (excluding obj/, bin/).
• If a .csproj, find its directory and scan .cs files within.
• If a .sln, parse it, find all project directories, and scan .cs files across all projects.

Always exclude obj/, bin/, and any user-specified exclusion patterns.

Step 2: Search for static dependency patterns

Scan each file for calls matching these categories:

Treat pattern matches as candidates, not findings. Before counting an instance call, trace how its receiver enters the class. A collaborator supplied through a constructor, parameter, property, or dependency injection (DI) is already a test seam. In particular, an injected HttpClient is testable with a controlled HttpMessageHandler; do not count its calls or recommend replacing it merely because the injected type is concrete.

For time calls, inspect use as well as count. Two ambient clock reads in one logical operation are two call sites and a consistency defect: for example, separate DateTime.UtcNow reads for CreatedAt and ExpiresAt = DateTime.UtcNow.AddDays(30) can drift. Recommend one captured instant. With TimeProvider, retain DateTimeOffset where possible; when the existing member requires UTC DateTime, use GetUtcNow().UtcDateTime, never .DateTime, which loses the UTC kind. Treat capturing one instant as an optional behavior-level follow-up: a mechanical wrapper migration must preserve the original reads one-for-one unless the user separately approves that semantic change.

Step 3: Aggregate and rank results

Count each call site across the entire scan scope — including the instance-member call sites covered by the rules below, not only static ones.

Counting rules — inaccurate totals are the main way this report loses to an ad-hoc scan:

• Build one occurrence ledger before writing prose. Give each included call site exactly one row containing category, exact pattern, file:line, and recommended seam. Derive every category, pattern, and per-file count by grouping that same ledger; never recount independently while writing tables.
• Keep the three count domains separate. Files scanned includes every eligible source file; affected files includes only files with ledger rows; call sites is the number of ledger rows. Never substitute one for another.
• One authoritative total. Every call site you found belongs in the category summary and the grand total. Never park real findings in an "additional observations" section that the totals exclude.
• Classify by what the member touches, not by whether it is static. Instance members that reach the same untestable resource still count and belong in the matching category (new FileInfo(path).LastWriteTimeUtc → File System; new HttpClient().GetAsync(...) → Network). Say "hidden dependency", not "static", when the member is an instance call.
• Check receiver provenance before counting instance calls. Count a resource access only when the code under test acquires or constructs the dependency itself. Exclude constructor-, parameter-, property-, and DI-injected collaborators from the "needs wrapping" total, including concrete HttpClient instances.
• Exclude deterministic pure helpers from the "needs wrapping" total. Path.Combine, Path.GetExtension, Path.GetFileName, and Math.*/string.* statics take no ambient input and are trivially testable. List them, if at all, in a separate "no action needed" note — never as testability blockers.
• Cover every category before reporting — time, file system, environment, network, console, process, randomness (new Random(), Guid.NewGuid()), culture (CultureInfo.CurrentCulture), and serialization/statics such as JsonSerializer. Omitting a category that is present is an under-count.
• Give file:line for every occurrence so the user can jump straight to it.
• Reconcile before publishing. The category totals, the top-patterns table, and the per-file table must sum to the same grand total.
• Treat exclusions as a scope decision, not a category. Remove obj/, bin/, generated, and user-excluded files before building the ledger. Do not include their files or call sites in any reported count. State the exclusions once rather than mixing excluded candidates into the arithmetic.
• Label truncated rankings. In a comprehensive audit, list all distinct patterns when needed for reconciliation. If the user asked only for a top-N subset, label it as a subset and do not imply that its rows sum to the grand total.

Produce a summary with:

1. Category summary — total call sites per category (time, filesystem, env, etc.)
2. Top patterns — the 10 most frequent individual patterns ranked by count
3. Most affected files — files with the highest number of static dependencies
4. Existing abstractions available — for each category, note the recommended .NET abstraction:
   • Time → TimeProvider (built-in since .NET 8)
   • File system → System.IO.Abstractions (NuGet package)
   • HTTP → IHttpClientFactory (built-in)
   • Environment → custom IEnvironmentProvider
   • Console → custom IConsole or ILogger
   • Process → custom IProcessRunner

Step 4: Present the report

Format the output as a structured report:

## Static Dependency Report

**Scope**: <project/solution name>
**Files scanned**: <count>
**Total static call sites**: <count>

### Category Summary
| Category     | Call Sites | Recommended Abstraction |
|-------------|-----------|------------------------|
| Time         | 42        | TimeProvider (.NET 8+) |
| File System  | 31        | System.IO.Abstractions |
| Environment  | 12        | IEnvironmentProvider   |
| ...          | ...       | ...                    |

### Top 10 Patterns
| # | Pattern             | Count | Files |
|---|---------------------|-------|-------|
| 1 | DateTime.UtcNow     | 28    | 14    |
| 2 | File.ReadAllText    | 18    | 9     |
| ...                                      |

### Most Affected Files
| File                          | Static Calls | Categories          |
|-------------------------------|-------------|---------------------|
| Services/OrderProcessor.cs    | 12          | Time, FileSystem    |
| ...                                                               |

### Migration Priority
1. **Time** (42 sites) — Use `TimeProvider`, zero NuGet dependencies on .NET 8+
2. **File System** (31 sites) — Use `System.IO.Abstractions` NuGet package
3. ...

Step 5: Suggest next steps

Based on the report, recommend which category to tackle first (highest count, best built-in support). Keep this to a few lines.

Mention generate-testability-wrappers or migrate-static-to-wrapper only when the user's next action clearly needs them — a hand-off note, not a sales pitch. Never end an audit with promotional next-steps that dilute the findings.

Validation

• All .cs files in scope were scanned (check count)
• Report includes category totals, top patterns, and affected files
• Category totals, top patterns, and per-file counts reconcile to the same grand total
• Files scanned, affected files, and call sites are reported as different quantities
• Every aggregate was derived from one occurrence ledger rather than independently recounted
• Every occurrence carries a file:line location
• No findings are held outside the totals in an "additional" section
• Calls on injected collaborators are excluded from the "needs wrapping" total
• Deterministic pure helpers (Path.Combine, Math.*) are not counted as testability blockers
• Each detected pattern has a recommended replacement listed
• obj/ and bin/ directories were excluded
• Migration priority is ordered by impact (count × ease of replacement)

Common Pitfalls