Skip to content

Multi-Agent KnowledgeOps Governance

This document defines how multiple AI agents can safely and consistently grow this repository over time without creating duplication, stale content, or low-confidence noise.

Goal

Build a durable documentation system where many agents can contribute in parallel while preserving:

  • Canonical ownership (one page per tool/topic)
  • Source traceability
  • Freshness and confidence signals
  • Reviewability through predictable PRs

Why this is the highest-leverage move

The main scaling risk is not "too little content", it is low-quality content growth. Without a shared operating contract, multiple agents eventually create duplicate pages, weak sourcing, and conflicting guidance. A common contract plus quality gates keeps throughput high and entropy low.

Multi-Agent KnowledgeOps Contract (Mandatory)

All AI-authored documentation PRs must satisfy the contract below.

  1. Respect canonical ownership.
  2. Before creating a page, search for existing tool/topic names and aliases.
  3. Update an existing canonical page when possible.
  4. Use repository templates and taxonomy.
  5. docs/templates/tool_template.md for tools/frameworks/providers.
  6. docs/templates/article_template.md for papers/articles.
  7. Place files in the taxonomy defined in docs/standards.md.
  8. Include auditable metadata in every AI-authored knowledge page update.
  9. Last reviewed date in ISO format (YYYY-MM-DD)
  10. Confidence level (high, medium, or low)
  11. Sources / References with at least one URL
  12. Limit each PR to one intent.
  13. Intake integration, curation pass, or audit fix.
  14. Avoid mixed PRs that combine unrelated tasks.
  15. Leave clear review context.
  16. State what was added, why it belongs, and what was deduplicated.

Role Model for Agents

Use role-specific behavior to reduce overlap and improve predictability.

Intake Agent

  • Scans sources and stages candidates in docs/new-sources.md
  • Proposes canonical destination and taxonomy tags
  • Does not perform broad refactors

Curation Agent

  • Integrates staged items into canonical pages
  • Normalizes structure to template and standards
  • Updates data/all_tools.json and mkdocs.yml when required

Audit Agent

  • Verifies metadata, links, and section completeness
  • Flags stale pages for refresh
  • Fixes low-risk quality issues in small PRs

CI Quality Gates

To make the contract enforceable, PR automation should check:

  1. Required metadata exists on changed knowledge pages.
  2. Sources / References exists and includes at least one URL.
  3. Confidence label is present and valid.
  4. Last reviewed date is valid ISO format.

These checks are implemented by scripts/check_docs_contract.py and run on pull requests.

Phased Rollout Plan

Phase 1: Contract and Structure

  • Publish this governance document.
  • Add contract language to docs/CONTRIBUTING.md.
  • Add metadata requirements to docs/standards.md.

Phase 2: Enforcement

  • Enable CI quality gate for changed Markdown docs.
  • Block merges when required metadata/sources are missing.

Phase 3: Reliability and Auditability

  • Add periodic audit runs for stale pages.
  • Track common failure modes and update agent prompts.

Definition of Done for AI-Authored PRs

A PR is complete only when:

  1. Target pages follow template/section expectations.
  2. Metadata and sources are present and valid.
  3. Canonical duplication checks were performed.
  4. Navigation/data indexes were updated when required.

Sources / References

Contribution Metadata

  • Last reviewed: 2026-02-26
  • Confidence: high