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.
- Respect canonical ownership.
- Before creating a page, search for existing tool/topic names and aliases.
- Update an existing canonical page when possible.
- Use repository templates and taxonomy.
docs/templates/tool_template.mdfor tools/frameworks/providers.docs/templates/article_template.mdfor papers/articles.- Place files in the taxonomy defined in
docs/standards.md. - Include auditable metadata in every AI-authored knowledge page update.
Last revieweddate in ISO format (YYYY-MM-DD)Confidencelevel (high,medium, orlow)Sources / Referenceswith at least one URL- Limit each PR to one intent.
- Intake integration, curation pass, or audit fix.
- Avoid mixed PRs that combine unrelated tasks.
- Leave clear review context.
- 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.jsonandmkdocs.ymlwhen 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:
- Required metadata exists on changed knowledge pages.
Sources / Referencesexists and includes at least one URL.- Confidence label is present and valid.
- 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:
- Target pages follow template/section expectations.
- Metadata and sources are present and valid.
- Canonical duplication checks were performed.
- Navigation/data indexes were updated when required.
Sources / References¶
- AI Hub Standards
- Contributing Guide
- Automated Contributions
- GitHub Actions: Events that trigger workflows
Contribution Metadata¶
- Last reviewed: 2026-02-26
- Confidence: high