PROVES Library: Agentic Knowledge for CubeSat Mission Safety
Automated dependency capture + knowledge graph + human review to prevent cross-team failures when critical knowledge disappears.
dY>y At a Glance
Capture
Extract dependencies with citations from docs, code, and issues.
Validate
Normalize into ERV relationships and check duplicates.
Review
Gate mission-critical items for human approval.
Store
Write entries to a structured knowledge graph.
Visualize
Render diagrams and trace transitive chains.
Learn
Use feedback to improve prompts and patterns.
dY?-‹,? Agentic AI Structure
%%{init: {'flowchart': {'defaultRenderer': 'elk'}}}%%
flowchart TB
subgraph Sources
A1[F Prime docs]
A2[PROVES Kit docs]
A3[Issues and PRs]
end
subgraph Ingestion
B1[Doc sync]
B2[Normalization]
end
subgraph Agents
C1[Curator agent]
C2[Extractor]
C3[Validator]
C4[Storage]
C5[Human review]
end
subgraph Knowledge Base
D1[Library entries]
D2[Knowledge graph]
D3[ERV relationships]
end
subgraph Outputs
E1[GitHub Pages diagrams]
E2[Queries and analysis]
E3[Risk scanner]
end
A1 --> B1
A2 --> B1
A3 --> B1
B1 --> B2
B2 --> C1
C1 --> C2
C2 --> C3
C3 -->|low/medium| C4
C3 -->|high/conflict| C5
C5 -->|approve| C4
C5 -->|reject| C1
C4 --> D1
C4 --> D2
D2 --> D3
D1 --> E1
D2 --> E2
D2 --> E3
dY` Lifecycle: Curation Run
sequenceDiagram
participant Doc as Documentation
participant Curator
participant Extractor
participant Validator
participant Review as Human Review
participant Storage
participant Graph as Knowledge Graph
Doc->>Curator: Ingest source
Curator->>Extractor: Capture ALL dependencies + citations
Extractor-->>Curator: Raw data with context
Curator->>Validator: Flag anomalies, check patterns
Validator-->>Curator: Staged with flags/notes
Curator->>Staging: Store ALL to staging tables
Staging-->>Review: Human verifies EVERY piece
Review-->>Graph: Only verified data enters truth
Graph->>Graph: Align sources, establish truth
Graph-->>Storage: Ack
Storage-->>Curator: Stored summary
dY”S Trial Results (What We Found)
- 45+ dependencies with citations across FA’ + PROVES Kit docs
- 4 cross-system dependencies not documented in either system
- 2 transitive dependency chains traced end-to-end
- 5 knowledge gaps that explain mission failures
dY”S Interactive Diagrams
- Dependency Overview
- Cross-System Dependencies
- Transitive Dependency Chains
- Knowledge Gaps
- Team Boundaries
dY”- What This Enables
- Detect risky changes before they cascade across subsystems.
- Preserve the “why” behind technical decisions.
- Enable cross-team learning without tribal knowledge loss.