API reference

trace_engine.apm.model.AttackPathModel

Core Attack-Path Model representing application architecture and attack surface.

find_paths_to_sinks(endpoint_id, cutoff=5, limit=25)

Find paths connecting an endpoint to any Sink node, capped at limit paths.

find_paths_from_endpoint(endpoint_id, max_depth=4, limit=25)

Find downstream paths from an endpoint up to max_depth, capped at limit paths.

summary()

Return a statistical overview of the graph.

trace_engine.ingest.repository.RepositoryScanner

Discovers and catalogs source code files across a project directory.

Only languages that a parser or adapter can use are ingested (ANALYZED_LANGUAGES). Markup, styles, data/config files and similar formats are skipped before they are opened, so they cost no I/O, hashing or graph nodes. Pass languages to widen or narrow the set.

scan()

Scans the repository and returns discovered source files.

trace_engine.security.fusion.NeuroSymbolicConfidenceEngine

Combines Neural (System 1 SecureBERT), Symbolic AST (System 2 Dataflow Sinks), and Dynamic Verification signals into mathematically calibrated Bayesian confidence.

fuse_evidence(evidence) classmethod

Calculates calibrated posterior probability from multiple independent security signals: Confidence = 1 - (1 - P_SecureBERT) * (1 - P_AST_Dataflow) * (1 - P_Dynamic_Oracle)

trace_engine.findings.model.Finding

Bases: BaseModel

Correlated vulnerability finding supported by both static AST and runtime proof.

trace_engine.harness.engine.AgentHarness

The TRACE Agent Harness provides an autonomous execution and verification environment

for AI coding models to inspect, patch, and mathematically prove security remediations.

get_next_target()

Returns the highest priority unmitigated finding packaged for agent consumption.

run_self_healing_loop()

Autonomous self-healing execution loop:

Iterates through prioritized findings, synthesizes defensive patches, applies them transactionally, and verifies runtime neutralization.