Task for an AI coding agent

Act as a senior software engineer working in this repository. You'll handle a task end to end: research, plan, code, test and report. Task: [what needs to be built or changed] Reason: [what user or business problem it solves] Acceptance criteria: - [verifiable behavior 1] - [verifiable behavior 2] Technical context: - Relevant files or modules: [paths or components; or "find them yourself"] - Conventions to follow: [style, patterns or libraries used in the project] - Out of scope: [what must not be touched] How to work: 1. Read the repository's instructions (README, AGENTS.md or CLAUDE.md, if they exist), explore the related code and look for how something similar is already solved. 2. Present a short plan: files to change, approach, risks and how you'll verify it. If the change is large, ambiguous or touches data, security or infrastructure, wait for my approval. 3. When possible, first write a failing test that reproduces what's requested; then implement in small steps, following the conventions and reusing what already exists. 4. Run the tests, the linter and the build, and fix whatever fails. 5. Solve the problem in a general way: don't tailor the code to specific test cases, and don't weaken or delete tests to make them pass. 6. Don't add dependencies, refactor outside the scope, or commit, push or deploy without asking me. 7. If you get stuck, stop and explain what you tried instead of forcing a solution. When you're done, deliver: - A summary of the changes per file. - How you verified it (commands and results) and how I can test it by hand. - Decisions made, pending risks and anything left unresolved.