From b63e24e0f045dbd34f418f9b6a7c992fdd1f0372 Mon Sep 17 00:00:00 2001 From: jsz Date: Thu, 29 Oct 2026 09:22:18 -0400 Subject: [PATCH 1/2] Add customer-impact threshold for review findings --- agents/AGENTS.md | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/agents/AGENTS.md b/agents/AGENTS.md index d28294e..3899fa0 100644 --- a/agents/AGENTS.md +++ b/agents/AGENTS.md @@ -48,12 +48,21 @@ When the target suite or Python version changes, update this section and - Do not use an em dash. Do not use `--` as punctuation. `--` belongs only in a CLI flag or a URL. - When the task is ambiguous, ask. Do not guess. - Build only what was asked. Do not add files, abstractions, dependencies, or features that were not requested. +- Do not design for rare or future cases nobody asked about. - Use the least code that solves the problem. After it works, do a second pass that simplifies and deletes unused code. - Before you hand work off, review your own output. Read the change as a critic against the request, the platform rule above, and the linked docs. Fix what that review finds, then hand it off. - Prefer boring, well-understood tools. If you reach for something novel, say why. - Match the style of the file you are editing. - Keep each pull request small enough to review. The size policy is in [PR Size and Scope](https://github.com/WLAN-Pi/developers/blob/main/CONTRIBUTING.md#pr-size-and-scope). +## Reviewing + +This section applies to agent reviewers, QA passes, and the self-review in How to work. + +- Judge customer impact before you raise an implementation gap. Describe a realistic task, the normal steps a user takes, and what the user materially loses. +- Report a finding only when it would materially prevent or mislead a customer doing ordinary work. Omit a low-impact finding. Do not include it at a lower priority. +- Rate each reported finding from 1 to 10 for likelihood and from 1 to 10 for severity. When likelihood times severity is below 5, open an issue and leave the code unchanged. A repository may set its own threshold in its `AGENTS.md`. + ## Python Follow [style/PYTHON.md](https://github.com/WLAN-Pi/developers/blob/main/style/PYTHON.md). From 315792af1a538b8ff9537945f73b17887672f15a Mon Sep 17 00:00:00 2001 From: jsz Date: Thu, 29 Oct 2026 11:13:25 -0400 Subject: [PATCH 2/2] Use an impact gate and labels for review findings --- agents/AGENTS.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/agents/AGENTS.md b/agents/AGENTS.md index 3899fa0..065c46c 100644 --- a/agents/AGENTS.md +++ b/agents/AGENTS.md @@ -59,9 +59,9 @@ When the target suite or Python version changes, update this section and This section applies to agent reviewers, QA passes, and the self-review in How to work. -- Judge customer impact before you raise an implementation gap. Describe a realistic task, the normal steps a user takes, and what the user materially loses. -- Report a finding only when it would materially prevent or mislead a customer doing ordinary work. Omit a low-impact finding. Do not include it at a lower priority. -- Rate each reported finding from 1 to 10 for likelihood and from 1 to 10 for severity. When likelihood times severity is below 5, open an issue and leave the code unchanged. A repository may set its own threshold in its `AGENTS.md`. +- Report a finding only if it would materially prevent or mislead a user doing ordinary work, or could cause data loss, a security hole, or a crash. Write the scenario first: the task, the normal steps, and what the user loses. If you cannot write a concrete scenario, drop the finding. +- After the scenario, label the finding with one word: Blocker (fix it in this change), Follow-up (file an issue and leave the code alone), or Drop (do not report it). A rare data-loss, security, or crash bug never drops below Blocker because it is rare. +- Do not report style, naming, docs, or rare future cases nobody asked about. A review with no comments is fine. ## Python