From 1b28115004e0dd897b3b1ae4b7064ee6d6590078 Mon Sep 17 00:00:00 2001 From: Dimitrij Drus Date: Fri, 6 Nov 2026 22:39:09 +0100 Subject: [PATCH] chore: Additional rules for agents --- AGENTS.md | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/AGENTS.md b/AGENTS.md index d66e66b..396ce96 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -14,6 +14,7 @@ This file provides repository-specific guidance for AI coding agents working on - Make the smallest coherent change. Do not include unrelated bug fixes, refactorings, dependency updates, or performance work; report those separately. For analysis-only requests, do not edit files or create patches unless asked. - Follow existing naming, formatting, file organization, error handling, and option patterns. Avoid unrelated formatting changes. - Apply SOLID pragmatically: keep responsibilities clear and abstractions justified by actual needs. Prefer readable, straightforward code (KISS before DRY); reuse existing mechanisms rather than introducing speculative interfaces or helpers. +- Keep validation close to the code responsible for the corresponding behavior. Separate static from context-dependent validation, and avoid redundant checks across layers. - Preserve public API compatibility unless explicitly asked otherwise, including option behavior, error contracts, and serialized representations. - Do not edit generated mocks manually or change `go.mod` / `go.sum` just to accommodate a local environment. - Preserve the library's framework independence and its `Message` abstraction. @@ -21,6 +22,7 @@ This file provides repository-specific guidance for AI coding agents working on ## RFC compliance and security - Check protocol-related changes against the relevant RFC text, not only existing tests or behavior on `main`. Explain consequential design decisions in terms of the applicable requirement. +- Distinguish RFC compliance issues from exploitable security vulnerabilities. Evaluate concrete threat scenarios, including risks introduced by stricter validation. - Validate before lossy conversion: preserve parameter presence, types, and values until the relevant checks are complete. - Respect the API contracts for Structured Fields component identifiers and any supported simplified configuration syntax. - Never relax signature, component, timestamp, nonce, algorithm, or key validation merely to satisfy tests. Assess security and compatibility implications explicitly. @@ -28,11 +30,12 @@ This file provides repository-specific guidance for AI coding agents working on ## Test design - Each unit test function should cover the meaningful positive and negative branches of its own target function or method. Higher-level tests should exercise that function's orchestration and error propagation, not duplicate detailed validation tests of its dependencies. +- When refactoring responsibilities, move the corresponding tests to the appropriate target functions and remove obsolete or overlapping scenarios. - Extend existing test functions where possible. Add only cases for new behavior, regressions, or genuinely uncovered edges; avoid overlapping scenarios. - Use table-driven tests only when they improve clarity. Keep cases independently understandable without unnecessary helpers or abstraction (KISS before DRY). - Put end-to-end signing, verification, and cross-API round trips in integration tests rather than repeating entire workflows in unit tests. - Assert observable contracts, including partial results and error identity or wrapping (`errors.Is` where appropriate), rather than incidental implementation details. -- Treat coverage as a diagnostic, not a target percentage. Inspect whether changed functions' meaningful branches have appropriate tests. +- Treat coverage as a diagnostic, not a target percentage. Inspect whether changed functions' meaningful branches have appropriate tests. Prefer removing unreachable or redundant code over adding artificial tests solely to increase coverage. - Use `integration_test.go` for cross-API workflows and `rfc_vectors_test.go` for RFC test vectors. ## Performance