Conversation
AGENTS.md should make using LLMs more effective by requiring less manual fixes of the generate code.
License Check Results🚀 The license check job ran with the Bazel command: bazel run //:license-checkStatus: Click to expand output |
|
The created documentation from the pull request is available at: docu-html |
|
Can you explain how to use this? AFAIK copilot will only pick up files like these automatically: https://github.com/eclipse-score/docs-as-code/tree/main/.github/instructions |
The world is spinning really fast. Copilot got support for The file was just created by typing |
|
This also states a top level |
AGENTS.md
Outdated
|
|
||
| The IPC boundary between gatewayd and someipd enforces **ASIL/QM context separation**—this is a critical architectural boundary. Changes affecting message transfer across this boundary require careful review. | ||
|
|
||
| **No direct ASIL/QM security features documented;** this is a QM-level project ([project_config.bzl](project_config.bzl)). |
There was a problem hiding this comment.
The gatewayd targets ASIL-B. Therefore the project is also ASIL-B, I guess?
There was a problem hiding this comment.
I have at the moment no idea how we deal with the situation that have both QM and ASIL-B in this repo. Anything related to vsomeip will be QM
There was a problem hiding this comment.
I removed this from the AGENTS.md file. I do not know what to do with project_config.bzl. This seems only to be required by Rust code and could get rid of it by removing the Rust code.
There was a problem hiding this comment.
so the agent will pick this even when i just tell it to do some random task inside this project . wont this create conflicts and add burden to the agent limited memory ? any prompt i give will make the agent cautious because of this critical stuff . this also looks vague to me and will cause the agent to hallucinate
"The IPC boundary between gatewayd and someipd enforces ASIL/QM context separation—this is a critical architectural boundary. Changes affecting message transfer across this boundary require careful review."
There was a problem hiding this comment.
yes it is used every time the agent is used. Actually the goal is to make it hallucinate less and save work for yourself by having less often to repeat yourself.
For example at the moment the agent needs to read the code to figure something out. If it would have learned the same from the AGENTS.md, it would not have needed to read the code and thus keep a smaller context window.
If we notice that instructions are misleading we can fix them.
We are ourselves confused how we will deal with that in this code
8b03d56 to
d22bd1e
Compare
|
Keep in mind there is some discussion in Slack about this PR and might be unlikely to be merged at the moment. |
AGENTS.md should make using LLMs more effective by requiring less manual fixes of the generate code.