Glossary

Technical, regulatory, and domain terms used in licit and its documentation.


Regulatory terms

TermDefinitionContext in licit
EU AI ActRegulation (EU) 2024/1689. The first comprehensive AI regulation worldwide. In force since August 2024, with phased enforcement through 2027.Regulatory framework evaluated by licit verify --framework eu-ai-act
FRIAFundamental Rights Impact Assessment. An impact assessment on fundamental rights required by Art. 27 of the EU AI Act for high-risk systems.Generated with licit fria
Annex IVAnnex IV of the EU AI Act. Defines the technical documentation required for AI systems: purpose, design, development, testing, performance.Generated with licit annex-iv
DeployerPerson or organization that deploys an AI system for use. Distinct from the provider who develops it.licit evaluates deployer obligations (Art. 26)
High-risk systemAI system classified as high risk under Annex III of the EU AI Act. Subject to additional compliance requirements.The FRIA is mandatory for these systems
OWASP Agentic Top 10List of the top 10 security risks for applications using AI agents (2025). Published by the OWASP Foundation.Framework evaluated by licit verify --framework owasp
NIST AI RMFAI Risk Management Framework (AI 100-1). NIST’s AI risk management framework.Planned for V1
ISO/IEC 42001International standard for AI management systems. Defines requirements for establishing, implementing, and improving AI management systems.Planned for V1
Compliance ratePercentage of evaluable requirements that are met. Formula: compliant / (total - n/a - not_evaluated) * 100Shown in reports and licit status

Technical terms

TermDefinitionContext in licit
ProvenanceOrigin and authorship of code. Tracking of which code was written by humans, generated by AI, or a combination.licit trace analyzes provenance
HeuristicRule based on observable signals (not certainties) to infer a classification. licit uses 6 weighted heuristics to classify commits as AI/human/mixed.Heuristics engine in provenance/heuristics.py
AttestationCryptographic integrity verification. In licit: HMAC-SHA256 for individual records, Merkle tree for batches.provenance.sign: true in config
Merkle treeBinary tree data structure where each node is the hash of its children. Allows verifying the integrity of data sets.Used to sign provenance batches
HMAC-SHA256Hash-based Message Authentication Code with SHA-256. Cryptographic signature that proves data has not been altered.Signing of individual provenance records
SARIFStatic Analysis Results Interchange Format. OASIS standard format for static code analysis results. Version 2.1.0.Read by VigilConnector
SBOMSoftware Bill of Materials. Formal inventory of software components. CycloneDX format used by vigil.Read by VigilConnector (sbom_path)
CycloneDXOWASP standard for SBOM. JSON/XML format for describing software components, their versions, and vulnerabilities.Supported SBOM format
JSONLJSON Lines. Format where each line is an independent JSON object. Ideal for append-only stores.Provenance store (.licit/provenance.jsonl)
Append-onlyStore that only allows adding records, never modifying or deleting them. Guarantees history immutability.ProvenanceStore

AI agent terms

TermDefinitionContext in licit
AI agentAI system that operates semi-autonomously or autonomously to perform tasks. In software development: Claude Code, Cursor, Copilot, Codex, architect.licit tracks code generated by agents
GuardrailControl that limits the behavior of an AI agent. Examples: protected files, blocked commands, code rules.Extracted from architect config, counts toward compliance
Quality gateAutomated check that must pass before accepting an agent’s output. Examples: lint, typecheck, test.Extracted from architect config
Human review gateCheckpoint that requires human approval. Example: branch protection with required reviews on GitHub.Detected from CI/CD config
Dry-runExecution mode that simulates actions without applying changes. Allows previewing the agent’s behavior.Detected from architect config
RollbackAbility to revert changes made by an agent. Typically via git revert or similar mechanisms.Detected from architect config
Session logRecord of an AI agent’s work session. Contains tools used, files modified, timestamps.Claude Code session reader
Budget limitSpending limit on tokens/API calls to control an agent’s resource consumption.Extracted from costs.budget_usd

Tool terms

TermDefinition
licitRegulatory compliance CLI tool for teams using AI in software development
EvidenceBundleDataclass with 18 fields that groups all evidence collected from the project
ProjectContextDataclass describing the auto-detected project context (languages, frameworks, CI/CD, etc.)
ControlRequirementAn individual requirement from a regulatory framework (e.g., Art. 9(1) of the EU AI Act)
ControlResultResult of evaluating a requirement against the project (status, evidence, recommendations)
GapItemIdentified compliance gap with actionable recommendation, estimated effort, and suggested tools
ConnectorRead-only integration with external tools (architect, vigil) that enriches evidence
ConnectorResultResult of a connector execution (files read, errors)
ComplianceFrameworkProtocol that defines the interface for regulatory framework evaluators
ConfigChangeDetected change in an AI agent configuration file, with MAJOR/MINOR/PATCH severity