doc: instruct firstmate to proactively state next steps - #2685
doc: instruct firstmate to proactively state next steps#2685indrasantosa wants to merge 2 commits into
Conversation
Captain feedback: after completing a step, say what happens next instead of going idle, and bundle a captain-hands-needed step with firstmate's own immediate follow-up so the captain isn't stuck relaying manually across turns.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e7479d6ae3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| **Keep the captain moving, not waiting.** | ||
| After completing a step toward a stated goal, say what happens next rather than going idle until asked. | ||
| When the very next step needs the captain's hands, such as a command only they can run because firstmate cannot write to a project, bundle that request with whatever firstmate itself will do right after, so one reply carries the whole next action instead of leaving the captain to relay it back manually. |
There was a problem hiding this comment.
Captain, when you have clearly and concretely approved a specific project edit, this example still says the command is one only you can run because Firstmate cannot write to a project. That contradicts the explicit exception permitting Firstmate to perform exactly such approved edits, so the new guidance can cause unnecessary manual handoffs; rephrase the example around an unapproved mutation or an interactive command while preserving the direct-operation exception.
AGENTS.md reference: AGENTS.md:L25-L27
Useful? React with 👍 / 👎.
| - A needed credential or login. | ||
|
|
||
| **Keep the captain moving, not waiting.** | ||
| After completing a step toward a stated goal, say what happens next rather than going idle until asked. |
There was a problem hiding this comment.
For a multi-step goal whose next operation is automatic or otherwise needs no captain action, this sentence requires a message after every completed step, directly conflicting with the existing instruction not to surface automatic fixes or routine progress. That ambiguity can turn internal execution into repeated progress updates; scope the announcement requirement to response-worthy stopping points or cases where work is actually waiting on the captain.
AGENTS.md reference: AGENTS.md:L476-L478
Useful? React with 👍 / 👎.
Scope the "state what happens next" guidance to response-worthy stopping points so it doesn't read as conflicting with "do not surface routine progress", and replace the project-write example with an unapproved-mutation/interactive-command example so it doesn't read as contradicting rule 1's captain-approved exceptions.
|
Speaking as Kun's firstmate: I reviewed the AGENTS.md next-steps instruction change and approved fork CI. Waiting on green checks including no-mistakes. |
2 participants
Summary
AGENTS.mdsection 9 (Escalation and captain etiquette): after completing a step toward a stated captain goal, proactively state what happens next rather than going idle, and bundle a step that needs the captain's own hands with whatever firstmate itself does right after.Test plan
bin/fm-doc-audience-check.shpasses (fm-doc-audience-check: ok surfaces=69 local_links=255)tests/fm-documentation-audiences.test.shpasses