Repository navigation
Conversation
## Overview `list_topics` answers could grow past `MaxToolResultBytes`. When that happened, the agent loop replaced the answer with "result too large". The model read that as a failed call and retried with a lower limit or switched tools, so one topic question took two or three calls. This PR makes `list_topics` size its own answer so it reaches the model. It also stops declining "how many users asked about this topic?", which `query_usage_by` can already answer. ## Changes - **Smaller topic rows** - Floats are rounded to what an answer would quote: share to 4 places, cost to 6, latency and trends to 1, cohesion to 3. - Examples are capped at 3 per topic (`topicsExamplesListed`). - `example_log_ids` is no longer sent. The `example_links` entries already carry the same ids, so `SampleLogIDs` is now `json:"-"`. - **Slimmer topics inside themes** - With `group_by theme`, each topic inside a theme is reduced to id, label, requests, share and link, because the theme already carries the counts, trend and series. - Each theme gets a `topic_count`, so the true size is still visible when fewer topics are listed. - The full topic rows remain available from `list_topics` without `group_by`. - **Fitting the result to the budget (`fitTopicResult`)** - If the encoded answer is still over `MaxToolResultBytes`, the smallest rows are dropped until it fits. The rows are ranked largest first, so this is an honest partial answer. - A `trimmed` note says what was cut and tells the model not to page through the rest. - Flat lists drop their tail topics. - Themed lists first shrink the topics inside every theme one rank at a time, then drop ungrouped topics, then drop the smallest themes. - Each pass re-encodes the answer. That costs a tool call, never a gateway request. - **Counting users behind a topic or theme** - The system prompt now tells the model to take the id from `list_topics` and call `query_usage_by` with `dimension user`, `topic_ids` set to that id, the same window and filters, and `limit 20`. - It counts rows other than `unassigned` and `other`, and says "at least 20" when 20 rows come back. - The model no longer offers the feature-request link for this question. - The tool result's `cannot_answer` note is replaced by a `people` note with the same instructions. - **E2E coverage** - The Postman `topics-people` case now expects `list_topics` followed by a `query_usage_by` call with `topic_ids` and `dimension: user`, and a numeric user count in the answer. It also expects no feature-request link, and the Topics folder description is updated to match. - The runner that builds the collection (`build-warp-collection.mjs`) has the same changes. ## Type of change - [x] Bug fix - [x] Feature - [ ] Refactor - [ ] Documentation - [ ] Chore/CI ## Affected areas - [x] Core (Go) - [ ] Transports (HTTP) - [ ] Providers/Integrations - [ ] Plugins - [ ] UI (React) - [ ] Docs ## How to test Run the Warp framework tests. `TestWarpListTopicsToolFitsTheToolResultBudget` uses realistic topics (uuid ids, four examples each, long labels, unrounded floats) and checks three cases: - The default call returns all 15 topics untrimmed, with example links and no `example_log_ids`. - `limit: 50` keeps the largest topics that fit and reports "N of the 50". - `group_by: theme` returns every theme with `topic_count`, and a `trimmed` note says the topics inside the themes were cut first. ```sh # Core/Transports go version go test ./framework/warp/... go test ./... ``` For the e2e case, rebuild the Warp Postman collection with `tests/e2e/api/runners/build-warp-collection.mjs` and run the Topics folder. "How many different users asked about the largest topic?" should call `list_topics` then `query_usage_by` (at most 3 calls) and answer with a user count instead of the feature-request link. No new configs or environment variables. ## Screenshots/Recordings Not applicable. ## Breaking changes - [ ] Yes - [x] No The tool-result shape changes for the model only: `example_log_ids` is removed, topics inside themes are slimmer, and `cannot_answer` is replaced by `people`. The log, store and HTTP APIs are unchanged. ## Related issues None linked. ## Security considerations None. The change only trims and rounds what `list_topics` returns. Scope and filter handling is unchanged, and the user count goes through `query_usage_by`, which already hides users the caller may not see behind `other`. ## Checklist - [ ] I read `docs/contributing/README.md` and followed the guidelines - [x] I added/updated tests where appropriate - [ ] I updated documentation where needed - [ ] I verified builds succeed (Go and UI) - [ ] I verified the CI pipeline passes locally if applicable
Contributor
|
Caution Review failedThe pull request is closed. Finishing Touches
Comment |
Contributor
 [Medium impact] The PR appears safe to merge; no actionable regressions were found. Reviews (1) · Last reviewed commit: "fixed pieline for arm build" · Reviewed by Greptile |
This branch is waiting to be deployed
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
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.
No description provided.