Knowledge Base

The Knowledge Base (also called the Vault) is Trinity's persistent memory system. It stores project documentation, architecture decisions, learnings from execution, and audit reports.

Accessing the Knowledge Base

Click Knowledge in the sidebar.

Structure

The knowledge base is organized hierarchically:

Books

Top-level containers. You'll typically see:

  • Your project name — project-specific knowledge
  • Gotchas — libraries of pitfalls (languages, frameworks, tools)
  • Trinity User Guide — this guide

Sections

Groupings within a book. For project knowledge, default sections include:

  • Roadmap — project vision, phases, milestones
  • Architecture — system design, data models, decisions, patterns
  • Operations — setup, running, deployment guides
  • Audits — codebase analysis reports
  • Per-target sections (e.g., "Web App", "API") with technical references

Chapters

Topic areas within a section. Chapters contain the actual content pages.

Pages

Individual knowledge articles written in markdown. Pages can be:

  • System-generated — created by Trinity during onboarding or execution
  • Agent-written — created by the Documenter agent after story execution
  • User-written — manually created or edited by you

How Knowledge Accumulates

During Onboarding

The onboarding wizard creates initial pages:

  • Roadmap overview, vision, phases, milestones
  • Architecture notes from your tech stack choices

During Import

The Import Existing flow generates:

  • Architecture documentation
  • Tech stack inventory
  • Feature catalog
  • Setup guide

During Execution

After each story, the Documenter agent writes:

  • What was built — summary of changes
  • Decisions made — why certain approaches were chosen
  • Discoveries — unexpected findings about the codebase
  • Patterns — reusable patterns identified during implementation

During Audits

Codebase analysis creates audit pages with:

  • Checklist items for improvements
  • Severity ratings
  • Recommendations

These checklists update automatically — when stories address audit items and merge, Trinity re-scans and checks off resolved items.

Searching

Use the search bar at the top of the Knowledge page to find content across all books:

  • Search by keyword
  • Results show matching pages with context
  • Click a result to navigate directly to the page

Agent Access

Agents have full access to the knowledge base during execution:

  • The Analyst reads relevant knowledge before planning
  • The Implementer consults architecture docs and patterns
  • The Auditor checks for known gotchas
  • The Documenter writes new entries

Agents access the vault through the /vault skill, which provides API endpoints for reading and writing.

The Audit Workflow

A powerful cycle connects audits to execution:

  1. Analyze — run a codebase analysis (via Import or manually)
  2. Audit pages created — checklists of improvements land in the vault
  3. Address items — click "Implement" from an audit page
  4. Stories generated — with audit_source linking back to the audit
  5. Execute — agents implement the fixes
  6. Auto-check — after merge, Trinity re-scans and checks off addressed items
  7. Re-check — manually trigger a re-check to verify remaining items

Exporting

Export knowledge as markdown for backup, sharing, or offline reference.

Zip Export

  1. Navigate to Knowledge
  2. Click the export button
  3. Download the generated zip file

The zip includes:

  • README.md — explains the structure and lists included books
  • Book metadata — _book.yml per book with name, description, icon
  • Section definitions — _sections.yml with chapter ordering
  • Chapter metadata — _chapter.yml per chapter with title and description
  • Pages — markdown files with YAML frontmatter (title, tags, source, timestamps)

PDF Export

Export formatted PDF documents from the knowledge base:

  1. Navigate to a book or section
  2. Click the PDF export button
  3. Choose scope (full book, section, or specific chapters)
  4. Download the generated PDF

PDFs include cover pages, metadata headers, and professional formatting.

Tips

  • Read before planning — review vault content before generating new PRDs to avoid duplicating existing knowledge
  • Trust agent entries — the Documenter writes factual, code-verified entries
  • Clean up periodically — remove outdated pages that no longer reflect the codebase
  • Use sections — organize content logically so agents can find what they need quickly