Repository Intelligence

Repository Intelligence is Ontoly's deterministic understanding of how a repository is structured and how it evolves over time.

It has two layers:

• structural repository intelligence from the compiler pass
• temporal repository intelligence from the History enhancer

Neither layer changes the Software Graph schema. Temporal intelligence is a derived artifact above the graph.

Structural Layer

The structural layer is implemented as a compiler pass:

createRepositoryIntelligencePass()

The pass reads repository configuration files from the source inventory and emits compiler symbols and relationships through normal graph construction.

Supported files include:

• package.json
• tsconfig.json
• pnpm-workspace.yaml
• turbo.json
• Dockerfile
• docker-compose.yml
• .env.example
• .github/workflows/*.yml
• Biome, ESLint, and Prettier configuration files

Structural nodes include Workspace, Package, Script, Dependency, Task, Pipeline, EnvironmentVariable, BuildTarget, Configuration, Container, Workflow, Job, Step, and Framework.

Structural relationships include CONTAINS, CONFIGURES, DEPENDS_ON, EXECUTES, PROVIDES, and USES.

Temporal Layer

The temporal layer is implemented by @0xsarwagya/ontoly-enhancer-history.

Input:

Repository
Git History
Software Graph

Outputs:

.ontoly/enhancers/artifacts/history.json
.ontoly/enhancers/artifacts/ownership.json
.ontoly/enhancers/artifacts/hotspots.json
.ontoly/enhancers/artifacts/cochanges.json
.ontoly/enhancers/artifacts/drift.json

The enhancer derives:

• first introduction commit
• last modification
• modification count
• contributors
• co-change relationships
• hotspot score
• churn score
• stability score
• ownership confidence
• bugfix, refactor, feature, and other commit ratios
• architectural drift timelines

The enhancer never reparses source files and never mutates SoftwareGraph.json.

History Availability

History collection is fail-closed. If Git history cannot be indexed, Ontoly returns a structured HISTORY_INDEXING_FAILED error instead of producing empty history artifacts. This prevents users and agents from mistaking unavailable history for evidence that no history exists.

Successful artifacts include:

{
  "historyIndexed": true,
  "historyStatus": {
    "indexed": true,
    "source": "git",
    "reason": null,
    "commitsCollected": 42
  }
}

The current Git collector uses a 128 MiB output buffer. Repositories whose Git history exceeds that limit fail explicitly rather than silently degrading.

CLI

ontoly history build .
ontoly history UserService
ontoly history feature "Authentication"
ontoly hotspots
ontoly ownership AuthService
ontoly churn
ontoly cochanges AuthService
ontoly stability AuthService
ontoly history validate --ci

ontoly history build writes all temporal artifacts. Query commands load those artifacts when present and can derive an in-memory history artifact from Git when the persisted artifact is missing.

MCP Capabilities

Repository Intelligence adds bounded MCP capabilities:

• History
• Ownership
• Hotspots
• Cochanges
• Stability

Responses are intentionally compact. They include graph provenance and deterministic confidence through the MCP response envelope.

Evidence Packs

When a History artifact is available, EvidencePack includes temporal sections:

• history
• ownership
• stability

These sections are attached only for the selected top graph nodes. Evidence Packs do not serialize the full Git history.

Invariants

• Git history is an input artifact, not compiler behavior.
• The Software Graph remains immutable.
• Outputs are deterministic for the same graph and Git history.
• Large commits are bounded when deriving co-change pairs.
• Ownership is repository ownership from Git history, not GitHub CODEOWNERS.
• No LLMs, embeddings, or vector search are used.