Add AGENTS.md to repo-smith#20
Conversation
|
@VikramGoyal23 Is there a reason for putting it in Ideally we would want our instructions to be AI-agnostic and not provide context for github copilot only right? |
|
@jovnc Forgive me, this draft is incomplete. Chat instructions are meant to be added in the |
|
@VikramGoyal23 Let's try to finish this by end of week |
|
@jovnc The file is ready for review. |
jovnc
left a comment
jovnc
left a comment
There was a problem hiding this comment.
Other than that LGTM, content-wise it's ok
There was a problem hiding this comment.
This AGENTS.md file would work if working in multi-agent environment
eg. core orchestrator that spawns sub-agents and delegates tasks to them
However, I'm not sure if it is supported by most agentic tools currently, eg. github copilot does not support this. This seems more suitable for tools like Claude Code.
AGENTS.md being a universal context should be independent of this, hence maybe it's better to phrase it another way than specifying different agents, eg. can use SKILL.md instead, or simply just state the responsibility of the agent when performing X task
|
@jovnc What do you think of this small extra comment? Also, I am slightly torn on the state of the file, since I feel as if this amount of context would be too much for a typical agent to have out the gate. |
SAN-MUYUN
left a comment
SAN-MUYUN
left a comment
There was a problem hiding this comment.
Other than the agent orchestration that you are trying to achieve here, the rest LGTM
| - **Imports**: Group in order: stdlib → third-party → local. Use absolute imports from `repo_smith` | ||
| - **Dataclasses**: Prefer `@dataclass` for simple data containers | ||
|
|
||
| Below is a list of agents you may be asked to act as. Check the relevant section for further context on how that particular agent would act. |
There was a problem hiding this comment.
This is a very good attempt, but let's not do Agent orchestration in this Agent.md. Agent orchestration should only be carried out when developer/contributor is fully aware of what each subagent is trying to do. This will cause huge amount of token to be used, and token usage by each subagent cannot be tracked easily.
3 participants
AGENTS.md provides instructions for AI coding agents working with the repo-smith codebase.