Conversation
nv-edwli
marked this pull request as ready for review
June 16, 2025 22:03
nv-edwli
marked this pull request as draft
June 16, 2025 22:03
nv-edwli
marked this pull request as ready for review
June 16, 2025 22:10
break cache Signed-off-by: Sean Lopp <slopp@nvidia.com>
break cache again Signed-off-by: Sean Lopp <slopp@nvidia.com>
hudalao
pushed a commit
to hudalao/aiq
that referenced
this pull request
Mar 10, 2026
Adds a 7th specialist to the deep researcher agent focused on the meta-level of research: searching for established frameworks, mental models, cross-domain analogies, design principles, and second-order implications rather than raw evidence. Changes: - New prompt: prompts/insight_synthesizer.j2 with a Research Protocol focused on framework search, analogy finding, design principle derivation, and scenario/what-if analysis - agent.py: register "insight-synthesizer" in specialist_configs and add "insight_synthesizer" to the prompt loading list; update class docstring and inline comments to reflect 7 specialists - researcher_coordinator.j2: list insight-synthesizer as sub-researcher NVIDIA-AI-Blueprints#7 and add dispatch rule for FRAMEWORKS / ANALOGIES / DESIGN PRINCIPLES questions https://claude.ai/code/session_01HZRynXmeNpvXWFMhA7bwFZ
ashan-nv
added a commit
that referenced
this pull request
Jul 1, 2026
…ch path Addresses two P1 review findings on the deep-research path for per-user MCP sources (e.g. Google Drive): #6 — validation ran before per-user MCP resolution. `validate_deep_research_tools` filtered a static, startup-built tool list; a per-user MCP source contributes no tools there, so a connected-GDrive-only deep-research request was rejected as "no tools available" and never submitted. It now treats a selected, configured per-user MCP source as a valid runtime tool candidate. Connectivity stays enforced by the submit preflight (evaluate_mcp_auth -> mcp_auth_required when not connected), and the authoritative tool check happens after the worker resolves tools. (Async path; the synchronous deep path still validates the static list and is a follow-up.) #7 — runtime MCP tools leaked into the orchestrator's callable catalog. Source tools were rendered into the orchestrator prompt's "Available Tools" section even though the orchestrator is bound only to helper tools + run_research_batch, so it called them directly and the runtime rejected them. The orchestrator prompt now advertises only its actually-callable tools, and a dedicated orchestrator middleware restricts the tool-name sanitizer allowlist to those tools. Source access stays delegated through run_research_batch to the researcher. Note: #7 changes the shared orchestrator prompt/middleware — recommend a deepresearch eval run before merge to confirm no quality regression. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Ashan Panduwawala <apanduwawala@nvidia.com>
ashan-nv
added a commit
that referenced
this pull request
Jul 1, 2026
…ch path Addresses two P1 review findings on the deep-research path for per-user MCP sources (e.g. Google Drive): #6 — validation ran before per-user MCP resolution. `validate_deep_research_tools` filtered a static, startup-built tool list; a per-user MCP source contributes no tools there, so a connected-GDrive-only deep-research request was rejected as "no tools available" and never submitted. It now treats a selected, configured per-user MCP source as a valid runtime tool candidate. Connectivity stays enforced by the submit preflight (evaluate_mcp_auth -> mcp_auth_required when not connected), and the authoritative tool check happens after the worker resolves tools. (Async path; the synchronous deep path still validates the static list and is a follow-up.) #7 — runtime MCP tools leaked into the orchestrator's callable catalog. Source tools were rendered into the orchestrator prompt's "Available Tools" section even though the orchestrator is bound only to helper tools + run_research_batch, so it called them directly and the runtime rejected them. The orchestrator prompt now advertises only its actually-callable tools, and a dedicated orchestrator middleware restricts the tool-name sanitizer allowlist to those tools. Source access stays delegated through run_research_batch to the researcher. Note: #7 changes the shared orchestrator prompt/middleware — recommend a deepresearch eval run before merge to confirm no quality regression. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Ashan Panduwawala <apanduwawala@nvidia.com>
ashan-nv
added a commit
that referenced
this pull request
Jul 1, 2026
…ch path Addresses two P1 review findings on the deep-research path for per-user MCP sources (e.g. Google Drive): #6 — validation ran before per-user MCP resolution. `validate_deep_research_tools` filtered a static, startup-built tool list; a per-user MCP source contributes no tools there, so a connected-GDrive-only deep-research request was rejected as "no tools available" and never submitted. It now treats a selected, configured per-user MCP source as a valid runtime tool candidate. Connectivity stays enforced by the submit preflight (evaluate_mcp_auth -> mcp_auth_required when not connected), and the authoritative tool check happens after the worker resolves tools. (Async path; the synchronous deep path still validates the static list and is a follow-up.) #7 — runtime MCP tools leaked into the orchestrator's callable catalog. Source tools were rendered into the orchestrator prompt's "Available Tools" section even though the orchestrator is bound only to helper tools + run_research_batch, so it called them directly and the runtime rejected them. The orchestrator prompt now advertises only its actually-callable tools, and a dedicated orchestrator middleware restricts the tool-name sanitizer allowlist to those tools. Source access stays delegated through run_research_batch to the researcher. Note: #7 changes the shared orchestrator prompt/middleware — recommend a deepresearch eval run before merge to confirm no quality regression. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Ashan Panduwawala <apanduwawala@nvidia.com>
AjayThorve
pushed a commit
that referenced
this pull request
Jul 2, 2026
…ch path Addresses two P1 review findings on the deep-research path for per-user MCP sources (e.g. Google Drive): #6 — validation ran before per-user MCP resolution. `validate_deep_research_tools` filtered a static, startup-built tool list; a per-user MCP source contributes no tools there, so a connected-GDrive-only deep-research request was rejected as "no tools available" and never submitted. It now treats a selected, configured per-user MCP source as a valid runtime tool candidate. Connectivity stays enforced by the submit preflight (evaluate_mcp_auth -> mcp_auth_required when not connected), and the authoritative tool check happens after the worker resolves tools. (Async path; the synchronous deep path still validates the static list and is a follow-up.) #7 — runtime MCP tools leaked into the orchestrator's callable catalog. Source tools were rendered into the orchestrator prompt's "Available Tools" section even though the orchestrator is bound only to helper tools + run_research_batch, so it called them directly and the runtime rejected them. The orchestrator prompt now advertises only its actually-callable tools, and a dedicated orchestrator middleware restricts the tool-name sanitizer allowlist to those tools. Source access stays delegated through run_research_batch to the researcher. Note: #7 changes the shared orchestrator prompt/middleware — recommend a deepresearch eval run before merge to confirm no quality regression. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Ashan Panduwawala <apanduwawala@nvidia.com>
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
3 participants
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.
Adds AIWB-specific specification files
Adds AIWB as an option under /deploy/