LearnNewsExamplesServices
Frontmatter
title>-
docsrestructure CodebaseOverview Query Entry Points to lead with
authortobiu
stateMerged
createdAtApr 12, 2026, 1:57 PM
updatedAtApr 12, 2026, 2:07 PM
closedAtApr 12, 2026, 2:07 PM
mergedAtApr 12, 2026, 2:07 PM
branchesdevagent/9900-docs-ask-knowledge-base
urlhttps://github.com/neomjs/neo/pull/9916
Merged
tobiu
tobiu commented on Apr 12, 2026, 1:57 PM

Resolves #9900.

Objective

The Codebase Overview (which is the foundational startup document for all agents) was structurally misaligned, disproportionately featuring query_documents over the strictly prioritized ask_knowledge_base tool.

Architectural Changes

  • Removed the implicit bias toward the legacy query_documents search workflow.
  • Placed ask_knowledge_base at the absolute forefront, codifying its role as the zero-cost RAG subagent for all conceptual discoveries.
  • Outlined explicit use-cases and example queries.
  • Added a structural warning against hallucinations via out-of-context training data.
  • Established a decision table (Ask -> Search -> Dissect).
tobiu
tobiu commented on Apr 12, 2026, 1:58 PM

Input from Antigravity (Gemini 3.1 Pro):

✦ # PR Review Summary

Status: Approved

Excellent work! The documentation now structurally prioritizes ask_knowledge_base, correctly enforcing the anti-hallucination workflow and zero-cost RAG subagent usage.


📊 Evaluation Metrics

  • [ARCH_ALIGNMENT]: 100 - Perfectly aligns with the AGENTS_STARTUP.md mandatory tool hierarchy and eliminates search bias.
  • [CONTENT_COMPLETENESS]: 100 - Included the decision table, example queries, and anti-pattern warning exactly as scoped.
  • [EXECUTION_QUALITY]: 100 - Clean markdown, correct PR formatting, and "Fat Ticket" conventions followed.
  • [PRODUCTIVITY]: 100 - Fully achieves the goals of Issue #9900.
  • [IMPACT]: 80 - Critical startup documentation that defines first actions for all new Swarm agents.
  • [COMPLEXITY]: 10 - Trivial file update, but highly strategic semantic value.
  • [EFFORT_PROFILE]: Quick Win - High ROI/Low Complexity. Minor text change with massive behavioral impact.

🕸️ Context & Graph Linking

  • Target Epic / Issue ID: Resolves #9900
  • Related Graph Nodes: ask_knowledge_base, Agent OS

🧠 Graph Ingestion Notes

  • [RETROSPECTIVE]: Agent documentation must always lead with vector-based RAG (ask_knowledge_base) rather than traditional file-search workflows (query_documents) to prevent multi-hop context window exhaustion. This PR solidifies that paradigm visually and structurally.

📋 Required Actions

To proceed with merging, please address the following:

  • All requirements met. Ready for Squash Merge.

Great job!


tobiu
tobiu commented on Apr 12, 2026, 2:06 PM

Input from Antigravity (Gemini 3.1 Pro):

✦ ### PR Update: Paradigm Rectification (Locked Columns)

Based on review feedback, it was identified that the example query: ask_knowledge_base(query='how to define a Grid with multiple bodies') promoted an anti-pattern (XY problem). In the target architecture, developers do not manually instantiate multiple grid bodies; they configure column locked: 'start' | 'end', and the core engine automatically constructs the required parallel DOM bodies under the hood.

Delta / Changes Made:

  • Modified CodebaseOverview.md (L551)
  • Replaced: ask_knowledge_base(query='how to define a Grid with multiple bodies')
  • With: ask_knowledge_base(query='how to configure locked columns in a Grid')

This ensures we don't accidentally train agents or developers to search for internal engine topology rather than the actual configuration API.