Improve plan/explorer delegation workflow and read-only allowlist#311
Open
Improve plan/explorer delegation workflow and read-only allowlist#311
Conversation
Tighten plan-mode Explore delegation guidance (single-call-first with clearer follow-ups) Allow Decide section in plan to ask questions Refine explorer/spawn prompt wording, and auto-allow common read-only investigation commands for plan & explorer agents.
zikajk
force-pushed
the
improve-plan-explorer-delegation
branch
from
fb910b9 to
df6a411
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Following up on our discussion regarding the Explorer subagent.
The current instructions have an issue where they conflict with the General agent’s instruction—specifically the part about 'researching complex questions.' The problem isn't just the conflict itself, but the fact that if the Explorer is preferred for deep research and then tries to use a skill intended to write findings to a file, it crashes. Because of this, I would still let the General subagent handle the vast majority of tasks.
The Explorer subagent makes sense in Plan mode, for example. Previously, the EXPLORE phase could eat up half the context window; that is now resolved (though there may still be rare cases where using a subagent isn't necessarily worth it or where model ignores instructions, but I’d say those will be minimal).
Without the Explorer, my plan consumed around 100k tokens. With this new delegated aggressive approach and the changes to Plan mode, I managed to get it down to 27k tokens. Personally, I see the Explorer’s greatest strength in how it handles browsing files, diffs, and similar tasks.
I’m curious to hear your thoughts on this. Give it a try, please:-)