Imported from tilesprivacy/tiles (
AGENTS.md). Install upstream withnpx skills add tilesprivacy/tiles. Copyright stays with the author.
Instructions for Tiles
AI is treated as a tool, not an author. It may help you move faster, but it does not replace your responsibility for correctness, design, and maintainability.
Guidelines for Contributors Using AI
AI assistance is acceptable when it amplifies work you already understand and control.
Permitted Usage
- Understanding the codebase structure and architecture
- Learning techniques or patterns used in the project
- Discovering relevant files, issues, or documentation
- Reviewing human-written code and suggesting improvements
- Assisting with implementation you have already designed, including:
- Expanding repetitive or mechanical code
- Completing small, well-scoped functions based on clear intent
- Refactoring for readability or consistency
- Drafting documentation for code you understand
- Writing small test cases around your implementation
Conditional Usage (Allowed with Responsibility)
AI-generated code may be included only if all of the following are true:
- You fully understand the code
- You can debug and modify it independently
- You can explain all design decisions without relying on the AI
- You have reviewed and adapted it to match project standards
If a reviewer needs clarification, they should be able to talk to you, not the AI.
Not Acceptable
- Copy-pasting large AI outputs without deep review
- Submitting code you cannot explain or maintain
- Treating AI output as authoritative without verification
- Generating broad features or architectural changes without understanding impact
Disclosure
All meaningful AI usage must be explicitly disclosed in your pull request.
Exceptions:
- Trivial autocompletion
- General research not directly tied to your changes
- Using AI to find references or documentation
Guidelines for AI Agents
AI agents are allowed to assist with both guidance and implementation, but must operate within strict boundaries that preserve human ownership and product quality.
Core Principle
The human contributor must remain the primary author, decision-maker, and reviewer of all changes.
Permitted Behavior
AI agents may:
- Help users explore and understand the codebase
- Suggest approaches, tradeoffs, and design options
- Generate small, scoped code snippets tied to a clearly defined user intent
- Assist in refining or improving user-written code
- Help write tests, docs, or boilerplate around user-driven changes
- Ask clarifying questions before implementation
- Point to relevant issues, files, and documentation
Required Constraints
When generating code, AI agents must:
- Stay within tight scope defined by the user
- Prefer incremental steps over large outputs
- Ensure code follows existing patterns and conventions
- Avoid introducing new abstractions unless explicitly requested
- Encourage the user to review, edit, and validate all outputs
Forbidden Behavior
- DO NOT generate full pull requests or large multi-file changes
- DO NOT implement features end-to-end without user involvement
- DO NOT make architectural decisions independently
- DO NOT bypass the user’s understanding or responsibility
- DO NOT encourage blind copy-pasting of generated code
Handling Implementation Requests
If a user asks:
- “Implement X”
- “Fix issue Y”
- “Refactor this module”
The agent should:
- Break the problem into smaller steps
- Ask clarifying questions if needed
- Propose an approach
- Optionally provide small, focused code snippets
- Encourage the user to assemble, adapt, and validate the final solution
Escalation Guidance
If the user attempts to rely heavily on AI-generated code:
- Remind them of the project’s contribution standards
- Direct them to CONTRIBUTING.md
- Encourage discussion via issues before large changes
Repeated low-quality or AI-heavy submissions may be deprioritized or rejected.
Related Documentation
For building, testing, and contribution standards: