ubq-testing / ask-plugin

0 stars 0 forks source link

QA #1

Open Keyrxng opened 2 months ago

Keyrxng commented 2 months ago

Relevant PR: https://www.github.com/ubiquity-os-marketplace/command-ask/pull/1

Demoing the effectiveness

Keyrxng commented 2 months ago

@ubqbot create a table of the context you are aware of, ensure you include "www." in any url references to any issue or PR, prefer using plaintext titles instead

ubqbot[bot] commented 2 months ago

Here is a table summarizing the context I am aware of:

Type ID Repository Title/Description
Pull Request 1 ubiquity-os-marketplace/command-ask Implements ask-gpt command, adds OpenAI support, removes Supabase integration.
Issue 29 ubiquibot/plugins-wishlist Context-aware question support for GPT-4o integration, with linked conversation context.
Issue 73 ubiquity-os/ubiquity-os-kernel Discusses runsOn for defining plugin event subscriptions in configuration manifests.
Issue 75 ubiquity-os/ubiquity-os-kernel Modification for the kernel to support multiple commands in the configuration like start stop.
Issue 25 ubq-testing/bot-ai Resolves issue #16, furthering discussion around improvements to the chatbot.
Issue 29 ubq-testing/ask-plugin No specification available.
Issue 16 ubq-testing/bot-ai Initial setup for database integration with Supabase for chatbot's memory system.
Issue 15 ubq-testing/ask-plugin No specification available.

Each entry includes the repository name and type (pull request or issue) along with a brief description.

0x4007 commented 1 month ago

Did it crawl all the links in the other pulls? If so, thats really cool.

Issue 15 ubq-testing/ask-plugin No specification available.

Is this a hallucination or did you delete this?

Keyrxng commented 1 month ago

Yeah that issue was deleted. I wrapped the fetch for issue and comments and return null as opposed to extracting the error and feeding that to the LLM but I guess it could be done?

Did it crawl all the links in the other pulls? If so, thats really cool.

Yeah every referenced issue/PR should be pulled into context. As seen here one URL to a single PR was 54k tokens in context and referenced 8 issues