Repository navigation
GLOSSARY
GitHub Action edited this page Oct 1, 2026
·
8 revisions
iso_doc_type: "Description" iso_viewpoint: "ComponentView" type: "module" title: "Glossary β Dark Gravity CA/CD Autonomous Factory" description: "ISO 42010 ComponentView / ISO 15289 Description documentation for Glossary β Dark Gravity CA/CD Autonomous Factory." tags: ['iso42010', 'okf'] timestamp: "2026-10-01T14:00:00Z" generated: "agent:okf-professional-documenter" verified: "true" last_verified_commit: "fbdc5848"
Purpose: Ubiquitous language dictionary for the Dark Gravity project. Defines all domain terms, agent names, protocol acronyms, and architectural concepts used throughout this wiki.
| Term | Definition |
|---|---|
| ADK | Agent Development Kit β Google's framework for building AI agents; used by AdkDriver to orchestrate multi-step agent reasoning |
| Aethelgard | Circuit breaker and anti-deadlock subsystem that monitors agent retry loops and trips after 3 consecutive failures with stagnant diff hashes |
| Agent | An autonomous software entity implementing the Agent trait (name() + execute()) that performs a specific role in the factory pipeline |
| Agent-Stuck | State entered when an agent's circuit breaker trips after exceeding the maximum retry count; requires HITL Vertex 3 (Architect Override) |
| AST | Abstract Syntax Tree β the parsed tree representation of Rust source code; used for OSR calculation and documentation symbol extraction |
| AuditorAgent | Factory agent responsible for security review, SAST scanning, and cryptographic credential verification |
| Term | Definition |
|---|---|
| Bounded Context | DDD concept defining clear boundaries between subsystems; Dark Gravity has 4: Agent Execution, Mission Orchestration, Infrastructure Integration, Security/Identity |
| Term | Definition |
|---|---|
| CA/CD | Continuous Agentic / Continuous Deployment β the autonomous software engineering paradigm where AI agents plan, code, test, and deploy with human governance gates |
| C4 Model | Architecture visualization framework with 4 levels: Context (C1), Container (C2), Component (C3), Code (C4) |
| Circuit Breaker | See Aethelgard; prevents infinite retry loops by tripping after 3 failed attempts with no code progress |
| CodeSurgeryExecutor | Trait in factory-core for applying surgical file patches (SurgicalPatch) and verifying syntax |
| CompliancePacket | Audit evidence record containing AST deltas, compute hours, FinOps spend, and Ed25519 NHI credentials |
| Term | Definition |
|---|---|
| DAG | Directed Acyclic Graph β the 6-phase mission execution graph orchestrated by Hatchet: Ingestion β Plan β Code β Validation β Review β Delivery |
| Dark Gravity | The CA/CD Autonomous Agent Factory project name |
| DDD | Domain-Driven Design β architectural methodology organizing code around business domains with explicit bounded contexts |
| DocAgent | DocumentationAgent β factory agent that generates wiki content, calculates OSR, and triggers doc-sync workflows |
| Term | Definition |
|---|---|
| Ed25519 | Elliptic curve digital signature algorithm used for Non-Human Identity (NHI) credential issuance and verification |
| ErrorCategory | Enum in factory-core classifying CI/CD pipeline errors: CodeCompilation, LintViolation, TestFailure, SecurityAudit, InfrastructureBuild, InfrastructureTransient, Unknown
|
| Term | Definition |
|---|---|
| FactoryError | Central error type in factory-core::error with variants: Config, Network, Internal, Security, Agent, Mission, Unexpected, Storage, IoError, RemediationError
|
| FinOps | Financial Operations β practice of tracking and optimizing cloud spend; implemented via FinOpsTag virtual headers and DailyBudgetConfig hardstops |
| FinOpsAgent | Factory agent tracking token consumption, budget enforcement, and spend velocity alerts |
| FinOpsTag | Struct with team, epic, microservice, environment, cost_center fields; serialized as x-vtags-* HTTP headers for LLM request attribution |
| FluxCD | GitOps controller that deploys the factory Kubernetes manifests from Git repositories |
| Term | Definition |
|---|---|
| gVisor | Google's container sandbox runtime providing kernel-level isolation for agent execution pods |
| GitOps | Infrastructure-as-code deployment methodology where Git is the single source of truth |
| Term | Definition |
|---|---|
| Hatchet | Workflow orchestration engine that executes the 6-phase mission DAG with step-level retry and timeout management |
| Hazitek 2026 | Basque Country R&D innovation grant program (via SPRI) funding the Dark Gravity project |
| HITL | Human-in-the-Loop β governance model ensuring humans retain oversight at 4 critical vertices |
| HITL Vertex 1 | Strategic Injection β Product Owner creates Epics tagged autonomous-plan
|
| HITL Vertex 2 | Sprint Mobilization β Tech Lead approves decomposed Spec-Kit tasks |
| HITL Vertex 3 | Exception Override β Architect resolves Agent-Stuck deadlocks |
| HITL Vertex 4 | Categorical Scrutiny β Senior Reviewer reviews design & merges PR/MR (agents strictly forbidden from merging) |
| Term | Definition |
|---|---|
| JitToken | Just-In-Time ephemeral security token with automatic memory zeroization (zeroize::ZeroizeOnDrop) |
| Term | Definition |
|---|---|
| Kafka KRaft | Apache Kafka running in KRaft mode (no ZooKeeper) for event streaming between factory subsystems |
| kube-rs | Rust Kubernetes client library used for pod management, sandbox orchestration, and cluster introspection |
| Term | Definition |
|---|---|
| LiteLLM | Multi-provider LLM proxy that routes inference requests to Azure OpenAI, Ollama, or other backends with unified API |
| Term | Definition |
|---|---|
| MCP | Model Context Protocol β JSON-RPC protocol for AI tool integration; the factory exposes tools via factory-mcp-server
|
| Mission | A complete unit of autonomous work: an issue or PR event that triggers a 6-phase DAG execution |
| MissionStatus | Enum: Pending, Running, Completed, Failed
|
| Term | Definition |
|---|---|
| NHI | Non-Human Identity β cryptographic Ed25519 credential proving agent authenticity; issued per mission and verified before code execution |
| Term | Definition |
|---|---|
| OKF | Open Knowledge Format β markdown documentation standard with YAML frontmatter, source path references, and architectural tags |
| Onion Architecture | Layered architecture where dependencies point inward: Domain (factory-core) β Application (factory-application) β Infrastructure (factory-infrastructure) β Interface (factory-mcp-server) |
| OpenZiti | Zero Trust overlay networking providing encrypted tunnels between factory services without traditional VPN |
| OSR | Orphan Symbol Rate β percentage of public Rust AST symbols not documented in the wiki; quality gate: OSR < 5% |
| Term | Definition |
|---|---|
| PipelineFailureEvent | Domain event representing a detected CI/CD pipeline failure from GitHub Actions or GitLab CI |
| PipelineScopeFilter | Utility evaluating whether a CI run is in-scope for remediation based on active PR/MR context |
| PolledIssueEvent | Domain event from the outbound poller representing a newly detected issue eligible for autonomous processing |
| PRDirective | Enum: Spec, Refine, Retry, Status, Validate, Interact β bot commands parsed from PR comments |
| PRCommentEvent | Domain event capturing a parsed @darkgravity or @dark-gravity PR comment with the extracted directive |
| Term | Definition |
|---|---|
| QAObserverAgent | Factory agent monitoring test coverage, quality metrics, and gate enforcement |
| Term | Definition |
|---|---|
| R2R (RAGtoRiches) | GraphRAG retrieval engine providing context-aware code search and knowledge retrieval for agent planning |
| Red-Green-Refactor | TDD discipline: write failing test (Red), implement to pass (Green), improve code structure (Refactor) |
| RemediationOutcome | Final result of a remediation attempt with status, fix PR URL, duration, and pipeline pass status |
| RemediationStatus | Enum: Pending, InProgress, Success, Failed, Escalated, Skipped
|
| RustantAgent | Planner agent executing the 6-phase Spec-Kit SDD sequence: init β specify β plan β execute β verify β git-commit |
| Term | Definition |
|---|---|
| SandboxConstraint | Struct defining pod resource limits: max_memory_mb (β€30), max_cpu_cores (β€0.25), network_egress_allowed (false) |
| SAST | Static Application Security Testing β automated code scanning via Semgrep; enforced as a gate before code execution |
| SastScanResult | Security scan output with score (0β10), is_safe, findings, critical_vulnerabilities_detected; passes gate when score β₯ 8.0 |
| SDD | Spec-Driven Development β the engineering methodology where specs, plans, and tasks are generated before code |
| SddMissionPlan | Structured plan output from RustantAgent containing parsed SDD tasks for Hatchet orchestration |
| SddTaskItem | Individual task from tasks.md with id, description, is_parallel, dependencies, target_files
|
| SecurityValidator | Trait in factory-core::security for Ed25519 signature validation and content auditing |
| Spec-Kit | The specification toolkit providing /speckit-specify, /speckit-plan, /speckit-tasks, /speckit-implement commands |
| SPRI | Sociedad para la PromociΓ³n y ReconversiΓ³n Industrial β Basque government agency administering Hazitek grants |
| SurgicalPatch | Struct for file-level code modifications: file_path, search_block, replace_block
|
| Term | Definition |
|---|---|
| Task | Individual unit of work within a Mission, with assigned_agent, dependencies, and status
|
| TaskStatus | Enum: Queued, Active, Finished, Blocked
|
| TDD | Test-Driven Development β methodology enforced by ZeroClawAgent: Red (write failing test) β Green (implement) β Refactor |
| Term | Definition |
|---|---|
| Vault | HashiCorp Vault β secret management service for API keys, tokens, and cryptographic material |
| Verification Triad | Three-gate quality pipeline: Logical (tests), Architectural (linting/deps), Security (SAST/NHI) |
| Term | Definition |
|---|---|
| ZeroClawAgent | Developer agent executing TDD tasks in gVisor-sandboxed pods with SAST pre-scan, checkpoint resumption, and circuit breaker integration |
| Zero Trust | Security model where no implicit trust is granted; all agent-to-service communication flows through OpenZiti encrypted tunnels |
| Zeroize | Rust crate ensuring sensitive data (JIT tokens, private keys) is cryptographically wiped from memory on drop |
- Business Context
- Strategic Design
- Tactical Design
- Agent Specifications
- Runtime Sequences
- Infrastructure Adapters
- Mission Data Model
- lib.md
- gitlab_verifier.md
- poller_service.md
- telemetry_export.md
- Agents
- Workflows
- Bridge
- Utils
- lib.md
- github.md
- gitlab.md
- jira.md
- kafka.md
- ziti.md
- vault.md
- r2r.md
- semantica.md
- aethalgard.md
- mcp_client.md
- pipeline_classifier.md
- s3.md
- sentry.md
- security_validator.md
- cursor_store.md
- git_poller.md
- lib.md
- main.md
- protocol.md
- sandbox.md
- scratch.md
- feedback_route.md
- github_webhook.md
- Skills
-
Tools
- mod.md
- bridge.md
- deep_research_tool.md
- execute_code.md
- get_factory_status.md
- index_code.md
- inspect_kafka_topic.md
- launch_sandbox_pod.md
- list_minio_buckets.md
- list_minio_objects.md
- plan_mission.md
- retrieve_context.md
- run_tests.md
- search_jira.md
- security_review.md
- spec_kit_tasks_to_issues.md
- spec_kit_tool.md
- update_mission_status.md