Agent answers off-domain questions instead of redirecting to LangChain ecosystem scope - #3
Merged
darieldatoon merged 1 commit intoJun 24, 2026
Conversation
…cosystem in seed AGENTS.md The seeded prompt framed the agent as general-purpose and forbade saying 'I don't know', causing it to confidently answer off-domain questions. Bound the identity to the LangChain ecosystem, add a Scope section that instructs out-of-scope redirects, and allow honest in-scope uncertainty.
|
LangSmith preview deployment was torn down.
|
darieldatoon
marked this pull request as ready for review
June 24, 2026 03:57
darieldatoon
deleted the
issues-agent/15fbae56-3a46-48b7-9d8c-039631e9e878
branch
June 24, 2026 03:57
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The agent answers questions clearly outside its LangChain/LangGraph/LangSmith scope — Django debugging, OAuth2 walkthroughs, Kubernetes/Jaeger architecture, CI/CD pipelines, transformer-attention theory, AI-startup business plans, cross-provider LLM comparisons — producing confident, detailed responses in domains it shouldn't claim authority over. The root cause is the seeded
AGENTS.mdin Context Hub repochat-lc-lite-agent-dariel(mirrored inutils/context_hub.py_SEED_AGENTS_MD), which frames the agent as a "general-purpose AI assistant with expertise in ... many other topics" and adds "Never say you don't know — give your best answer," with no scope-restriction or redirect clause. For a domain-scoped assistant this is a fundamental UX problem on every off-topic request.