Framework Glossary

This glossary defines the key terms introduced in the handbook. Each entry includes a short definition, a link to the full glossary entry on the site, and a reference to the handbook page where the concept is explained in detail.

Terms

AgentOps Dashboard — The real-time monitoring interface that tracks agent execution status, pipeline health, and compute costs across a squad. The Flow Manager reviews it during the Daily Flow Sync. Handbook: Ceremonies and Routines

Blocker Flag — A signal raised by an agent when it encounters ambiguity or a constraint it cannot resolve, triggering escalation to a human operator who executes a Rescue Mission. Handbook: The Orchestration Layer

Context Engineering — The practice of designing, organizing, and managing the information that flows into an LLM's context window. Takes a systems-level view of how system prompts, documents, history, and tools are assembled and prioritized. Handbook: Context Management

Context Index — The structured knowledge base that agents draw from during execution. Contains Live Specs, architectural rules, Golden Samples, domain glossaries, and historical context, organized for machine consumption. Handbook: Context Management

Context Packet — A bundled collection of specifications, architectural rules, Golden Samples, and domain context assembled for a specific task. Created during weekly Specification Engineering Blocks and consumed during daily agent execution. Handbook: Ceremonies and Routines

Continuous Development Loop — The agentic evolution of CI/CD that automates the entire lifecycle from spec to production. Extends the traditional pipeline with spec injection, automated context assembly, Eval Harness gates, and knowledge feedback loops. Handbook: From Agile to Agentic

Core Nucleus — Code that is too architecturally sensitive or security-critical for agent execution. Written exclusively by human Agent Operators, it includes authentication paths, core algorithms, and foundational abstractions. Handbook: The Hybrid Squad

Evaluation Harness — The automated test suite that validates every agent output before it reaches a human reviewer. Combines functional tests, security scans, architectural conformance checks, and LLM-as-a-Judge evaluations. Handbook: Evaluation Harness

Golden Samples — Curated reference implementations that agents use as templates for new code. A well-chosen golden sample teaches an agent more than pages of written instructions by demonstrating expected patterns through concrete example. Handbook: The Hybrid Squad

Human-in-the-Loop — A design principle where human judgment is required at critical decision points before an agent proceeds with consequential actions. Enables graduated autonomy through tiered permission models. Handbook: The Orchestration Layer

Live Spec — A machine-readable, deterministic contract that defines what to build, why it matters, and how to verify it works. Contains a Behavioral Contract, System Constitution, and Actionable Task Map. Evolves alongside the codebase under version control. Handbook: Spec-Driven Development

Rescue Mission — The intervention workflow where an Agent Operator diagnoses a stuck agent, injects missing context, and resumes execution. Follows a three-step process: Diagnose, Inject, Resume. Handbook: Ceremonies and Routines

Spec-Driven Development — The practice of replacing informal user stories with machine-readable specifications that serve as deterministic contracts between humans and agents. Makes the core pillars operational by providing precise inputs for context, governance, and routing. Handbook: Spec-Driven Development

Token Budget — The maximum compute spend authorized per time period, functioning as a hard constraint that prevents runaway agent loops. Enforced at the Orchestration Layer and tracked on the AgentOps Dashboard. Handbook: Ceremonies and Routines