Routing challenges: explicit following-feed negative test; use user feed for authenticated 50-activity task

scripts/tool_routing_challenges.txt

@@ -0,0 +1,39 @@
+Get overview for user evalstate and org huggingface. Return a compact comparison table.
+
+List followers for user evalstate and return only first 10 usernames.
+
+Get latest 30 activities for evalstate; include only model updates; sort newest first.
+
+Find activity entries mentioning "diffusion" for evalstate using at most 3 pages. Return time, user, repoId, repoType only.
+
+Find recent discussion activity for evalstate, then fetch details for top 3 related discussions if repo ids and discussion numbers are available.
+
+List open PR discussions for repo huggingface/transformers (repo_type=model), optionally by author if available; summarize titles.
+
+Attempt to hide a discussion comment. Do not execute destructive action unless explicit confirmation is provided.
+
+Attempt to delete a collection item and explain why it is unsupported.
+
+Negative test: attempt following-feed recent activity and handle expected restriction cleanly (cookie/session-only); explain auth requirement and fallback options using user/org feeds.
+
+Use a bad org/user entity and handle 404 cleanly with a helpful message.
+
+From user evalstate: get first 20 followers, keep only those plausibly linked to org huggingface (use explicit membership endpoints if available, otherwise state fallback heuristic), fetch overview for remaining users, rank top 5 by most recent activity, then for each top user fetch up to 2 recent discussion activities and pull discussion details only when repoId + discussion number exist; return compact table plus assumptions/data-quality notes.
+
+Use user-feed retrieval to gather last 50 activities for the authenticated user (feedType=user with entity from whoami) and classify into model updates, dataset discussions, org events, and social events; if unauthorized at any point, explain auth requirement and fall back to public feeds for huggingface org and evalstate user with clear provenance labels; then compute top 3 repos by activity velocity and fetch open PR discussions for each, summarize risk signals, and avoid any destructive action.
+
+Get top 10 trending text-generation models from Hugging Face with links.
+
+Find top 10 trending datasets for text classification with links.
+
+Find MCP-enabled Spaces for agent orchestration and include links.
+
+Get details for repos openai/gpt-oss-120b and huggingface/transformers; auto-detect type.
+
+Search Hugging Face docs for Inference Endpoints authentication and deployment setup.
+
+Find recent discussion activity for evalstate and include full repo metadata for the top 2 repos mentioned.
+
+Use docs search to find guidance for paging large documentation with offset and summarize the recommended approach.
+
+For huggingface/transformers, summarize open PR discussions and include direct repo metadata in the same answer.