Skip to content

Commit 3907ae8

Browse files
samtrionsamtrion
committed
chore(template): sync with dailydevops/dotnet-template [skip ci]
1 parent 9da992c commit 3907ae8

File tree

1 file changed

+164
-0
lines changed

1 file changed

+164
-0
lines changed

‎decisions/2025-11-01-ai-agent-build-restore-test-permissions.md‎

Lines changed: 164 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,164 @@
1+
<!-- List of authors who contributed to this decision. Include full names and roles if applicable. -->
2+
authors:
3+
- Martin Stühmer
4+
5+
<!--
6+
The patterns this decision applies to. Each entry is a glob pattern that matches files affected by this decision.
7+
-->
8+
applyTo:
9+
- "**/*.csproj"
10+
- "**/*.fsproj"
11+
- "**/*.vbproj"
12+
- "**/*.sln"
13+
- "**/*.slnx"
14+
- "**/tests/**/*.cs"
15+
- "AGENTS.md"
16+
- ".github/copilot-instructions.md"
17+
18+
<!-- The date this ADR was initially created in YYYY-MM-DD format. -->
19+
created: 2025-11-01
20+
21+
<!--
22+
The most recent date this ADR was updated in YYYY-MM-DD format.
23+
IMPORTANT: Update this field whenever the decision is modified.
24+
-->
25+
lastModified: 2025-11-01
26+
27+
<!--
28+
The current state of this ADR. If superseded, include references to the superseding ADR.
29+
Valid values: proposed, accepted, deprecated, superseded
30+
-->
31+
state: accepted
32+
33+
<!--
34+
A compact AI LLM compatible definition of this decision.
35+
This should be a precise, structured description that AI systems can easily parse and understand.
36+
Include the core decision, key rationale, and primary impact in 1-2 concise sentences.
37+
-->
38+
instructions: |
39+
AI agents are authorized to execute dotnet build, restore, and test commands for all projects in the solution to verify code correctness and maintain quality standards.
40+
All operations MUST be executed at the solution level using the .sln or .slnx file found in the repository root directory.
41+
Before committing changes, AI agents MUST run restore, build, and test operations to ensure no regressions are introduced.
42+
---
43+
# Decision: AI Agent Authorization for Build, Restore, and Test Operations
44+
45+
AI coding assistants such as GitHub Copilot are authorized to execute `dotnet build`, `dotnet restore`, and `dotnet test` commands for all projects within the solution to verify code correctness, ensure compilation success, and validate functionality through automated tests.
46+
47+
## Context
48+
49+
AI coding assistants are increasingly capable of making complex code changes across multiple files and projects. However, without the ability to verify their changes, several issues arise:
50+
51+
- **Undetected Compilation Errors**: Code changes may introduce syntax errors, missing references, or type mismatches that prevent successful compilation.
52+
- **Broken Dependencies**: Package references may be incomplete or incorrect, causing restore failures.
53+
- **Regression Bugs**: Changes may break existing functionality that is covered by automated tests.
54+
- **Integration Issues**: Multi-project changes may compile individually but fail when integrated.
55+
- **Quality Degradation**: Without verification, the quality of AI-generated code cannot be assured before commit.
56+
- **Developer Burden**: Human developers must manually verify all AI-generated changes, reducing productivity gains.
57+
58+
To maximize the effectiveness of AI coding assistants while maintaining code quality, agents need the capability to validate their own work through standard .NET development operations.
59+
60+
## Decision
61+
62+
We authorize AI coding assistants to execute the following .NET CLI commands:
63+
64+
1. **`dotnet restore`**: Restore NuGet packages and dependencies for projects and solutions.
65+
2. **`dotnet build`**: Compile projects and solutions to verify code correctness and identify compilation errors.
66+
3. **`dotnet test`**: Execute unit tests, integration tests, and other automated tests to validate functionality.
67+
68+
### Implementation Guidelines
69+
70+
AI agents MUST:
71+
72+
- Detect the solution file (`.sln` or `.slnx`) in the repository root directory and use it for all build, restore, and test operations.
73+
- Execute all commands at the solution level to ensure consistency across all projects and detect integration issues.
74+
- Execute `dotnet restore` before building if package references have been added or modified.
75+
- Execute `dotnet build` after making code changes to verify compilation success.
76+
- Execute `dotnet test` after code changes that may affect functionality to ensure no regressions.
77+
- Report compilation errors, warnings, and test failures clearly to the user.
78+
- Iterate on fixes when errors or test failures are detected.
79+
- Use appropriate command-line options (e.g., `--configuration Release`, `--no-restore`) as context requires.
80+
81+
AI agents SHOULD:
82+
83+
- Use `--verbosity` options appropriately to provide useful diagnostic information.
84+
- Consider build performance implications and avoid unnecessary rebuilds.
85+
- Verify that the solution file exists before executing commands and report an error if missing.
86+
87+
AI agents MAY:
88+
89+
- Execute `dotnet clean` before building to ensure a clean build environment.
90+
- Use additional CLI options (e.g., `--framework`, `--runtime`) when targeting specific configurations.
91+
- Run specific test filters (e.g., `--filter Category=Unit`) when appropriate.
92+
93+
## Consequences
94+
95+
### Positive Consequences
96+
97+
- **Improved Code Quality**: AI-generated code is validated before being presented to developers.
98+
- **Faster Development Cycles**: Errors are caught and fixed immediately by the AI agent rather than discovered later by developers.
99+
- **Reduced Human Oversight**: Developers can trust that AI-generated changes compile and pass tests.
100+
- **Better Error Context**: AI agents can see and respond to actual compiler and test errors rather than relying on static analysis alone.
101+
- **Continuous Verification**: Changes are validated in real-time as part of the AI's workflow.
102+
- **Test-Driven Development Support**: AI agents can run tests iteratively while implementing features.
103+
104+
### Potential Negative Consequences
105+
106+
- **Resource Usage**: Build and test operations consume CPU, memory, and disk I/O resources.
107+
- **Time Overhead**: Each verification cycle adds time to AI operations.
108+
- **Potential for Build Loops**: Poorly designed fixes could lead to repeated failed build attempts.
109+
- **Complexity**: AI agents need to interpret and respond appropriately to build and test output.
110+
111+
### Mitigation Strategies
112+
113+
- Use incremental builds where possible to minimize resource usage.
114+
- Set reasonable timeout limits for build and test operations.
115+
- Implement clear error handling and reporting mechanisms.
116+
- Monitor for and prevent infinite fix-attempt loops.
117+
- Allow users to disable automatic verification if needed.
118+
119+
## Alternatives Considered
120+
121+
### 1. Manual Verification Only
122+
123+
**Description**: Require human developers to manually run build and test commands after AI-generated changes.
124+
125+
**Why Not Chosen**:
126+
- Reduces the value proposition of AI assistance
127+
- Increases developer workload and slows down development
128+
- Errors are discovered later in the workflow
129+
- Does not leverage AI's ability to iterate on fixes
130+
131+
### 2. Static Analysis Only
132+
133+
**Description**: Use static code analysis and linting without actual compilation or test execution.
134+
135+
**Why Not Chosen**:
136+
- Cannot detect runtime issues or test failures
137+
- May miss compilation errors not caught by static analysis
138+
- Less comprehensive verification than actual build/test
139+
- Doesn't validate package dependencies and restore operations
140+
141+
### 3. Sandbox/Mock Build Environment
142+
143+
**Description**: Create a separate, isolated environment for AI build verification that doesn't use actual project files.
144+
145+
**Why Not Chosen**:
146+
- Adds significant complexity to implementation
147+
- May not accurately reflect actual build behavior
148+
- Difficult to maintain parity with real build environment
149+
- Adds maintenance overhead for separate build configuration
150+
151+
### 4. Read-Only Analysis Without Execution
152+
153+
**Description**: Allow AI agents to analyze code but not execute any commands.
154+
155+
**Why Not Chosen**:
156+
- Severely limits AI effectiveness
157+
- Cannot verify changes work correctly
158+
- Misses the primary benefit of automated verification
159+
- Contradicts the goal of autonomous AI assistance
160+
161+
## Related Decisions
162+
163+
- [Centralized Package Version Management](./2025-07-10-centralized-package-version-management.md) - AI agents must respect centralized package versions when executing restore operations
164+
- [.NET 10 and C# 13 Adoption](./2025-07-11-dotnet-10-csharp-13-adoption.md) - Build operations must target the correct .NET version and C# language version

0 commit comments

Comments
 (0)