fix(discover): stop telling the agent to filter when the content filter is off - #384
Merged
Merged
Conversation
…er is off The content filter switch only skipped DeviceSafetyFilter on the parsed answer; the system prompt always carried the anti-piracy guardrail and BLOCK rules, so the model refused before the filter was ever reached. The prompt now follows DiscoverQuery.contentFilter: with it off, the guardrail is left out and risky links are ranked lower and explained instead of dropped.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![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: a5706a1a05
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".
Contributor
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
With Content filter switched off, Discover still refused requests such as "3 hottest tv shows", with a Filtering step that cites an anti-piracy policy. The switch only skipped
DeviceSafetyFilteron the parsed answer. The system prompt was the same either way: it always carried the ANTI-PIRACY GUARDRAIL and theBLOCK …device-safety rules, so the model refused before the filter was ever reached.Change
ResourceDiscoveryService.SYSTEM_PROMPTis nowsystemPrompt(contentFilter), and each run passes inDiscoverQuery.contentFilter.BLOCK …rules become "rank riskier links lower and say why indeviceSafetyNotes", and the output rule says "If no candidates" instead of "If no safe candidates". The model still prefers official sources, and the rules about ignoring instructions in fetched pages and never downloading automatically stay.docs/ai-discovery.mdandAGENTS.mddescribe the new behavior. The provider's model may still decline a request on its own.Testing
BLOCKrules with the filter on and has neither with it off../gradlew :ai:discover:testpasses.