The glossary · Agentic Engineering

Category

126 of 126 terms

Agent: Concepts; A model that decides its own next step and calls tools in a loop until a goal or budget stops it.
Agent loop: Concepts; Read context, propose a tool call, run it, append the result, repeat.
Agentic engineering: Concepts; Karpathy's February 2026 term for professional agent-driven development: you orchestrate agents that write most code and provide oversight; engineering because it takes expertise.
Agentic search: Patterns; Iterative searching with tools such as grep and file reads instead of a single retrieval step.
AGENTS.md: Coding agents; An open Markdown convention for project instructions read by many coding agents; released by OpenAI in 2025 and stewarded by the Agentic AI Foundation.
A2A: Protocols and ops; Agent2Agent protocol for agents from different systems to communicate and delegate.
Backoff: Protocols and ops; Waiting longer between retries, usually doubling each time, to recover from rate limits.
Batch API: Protocols and ops; Asynchronous bulk requests at lower cost, for work that is not urgent.
Brief: Patterns; The prompt a parent agent writes for a subagent; everything the subagent will know.
Budget: Concepts; A hard cap on steps, tokens, time or cost for one run.
Bypass permissions: Coding agents; A Claude Code permission mode that skips approval prompts; use only inside a sandbox.
Checkpoint: Coding agents; A saved state of the session and files that you can rewind to.
CLAUDE.md: Coding agents; Project instructions Claude Code loads at session start.
Claude Code: Coding agents; Anthropic's agentic coding tool for the terminal, IDE, desktop and web.
Chunking: Patterns; Splitting documents into retrievable pieces for RAG.
Citations: Patterns; Links from claims to the retrieved sources that support them.
CLI vs MCP: Patterns; CLI when a shell and a well-known tool exist; MCP for no-shell clients, managed auth and SaaS.
Codex: Coding agents; OpenAI's coding agent, available as a CLI and in the cloud.
Compaction: Concepts; Summarizing older conversation to free context; lossy by design.
Context engineering: Concepts; The set of strategies for curating and maintaining the optimal set of tokens during inference (Anthropic, September 2025).
Context rot: Concepts; Declining recall and accuracy as the number of tokens in the context window grows.
Context window: Concepts; Everything the model can see in one call, measured in tokens.
Cost per accepted task: Protocols and ops; Total spend divided by the number of runs whose output was accepted.
Depth limit: Patterns; A cap on how many levels agents may delegate to other agents.
Embeddings: Patterns; Vectors that represent the meaning of text, used for similarity search.
Errors as observations: Patterns; Returning tool failures to the model as results so it can correct itself.
Eval: Concepts; A repeatable test that scores model or agent output over a set of cases.
Evaluator-optimizer: Patterns; One agent produces, another critiques, repeat until good enough.
Exit code 2: Coding agents; In a Claude Code hook, a blocking error: PreToolUse blocks the call and stderr goes to Claude.
Extended thinking: Concepts; The model spends extra tokens reasoning before answering.
Fan-out / fan-in: Patterns; Running independent subtasks in parallel and combining the results.
Function calling: Concepts; Another name for tool use: the model emits a structured call for your code to run.
Golden set: Protocols and ops; A fixed set of real inputs with known good outputs, used for evals.
Grounding: Concepts; Tying model output to retrieved facts or sources.
Guardrail: Concepts; A check enforced in code that blocks or corrects unsafe or invalid actions.
Hallucination: Concepts; Confident output not supported by facts or sources.
Handoff: Patterns; Passing a task and its context from one agent to another.
Harness: Concepts; Everything around the model that makes it an agent: prompt, tools, permissions, hooks, memory, orchestration.
Harness engineering: Concepts; Designing the guides and sensors that wrap a coding agent to increase confidence in its output; named as a discipline in early 2026.
Headless mode: Coding agents; Running Claude Code non-interactively, for example claude -p, in CI or scripts.
Hook: Coding agents; A command Claude Code runs on a lifecycle event; it can block, inject context or log.
HTTP+SSE transport: Protocols and ops; MCP's original remote transport, replaced by Streamable HTTP and now deprecated.
Human in the loop: Patterns; A person approves at defined checkpoints where mistakes are expensive.
Human review gate: Patterns; A pause in a pipeline that waits for approval before an expensive or irreversible step.
Hybrid search: Patterns; Combining keyword and vector search for retrieval.
Just-in-time context: Concepts; Loading rules or documents at the moment they matter instead of up front.
Latency: Protocols and ops; Time to first token and total time to complete a response or run.
Least privilege: Concepts; Giving an agent only the tools and access the task needs.
Lethal trifecta: Concepts; Private data, untrusted input and an exfiltration path in one agent; enables data theft by injection.
LLM-as-judge: Protocols and ops; Using a model to grade outputs against a rubric.
MCP: Protocols and ops; Model Context Protocol: an open standard that exposes tools, resources and prompts to agents.
MCP prompts: Protocols and ops; Reusable prompt templates an MCP server offers to users.
MCP resources: Protocols and ops; Read-only data an MCP server exposes to the model.
MCP tools: Protocols and ops; Actions an MCP server lets the model call.
Memory: Concepts; Facts persisted across sessions, usually in files the agent reloads.
Model routing: Patterns; Sending each request to a specific model or specialist based on its type.
Model tiering: Patterns; Using strong models for planning and review, cheaper ones for execution.
Non-determinism: Concepts; The same input can produce different outputs.
Observability: Protocols and ops; Metrics, logs and traces across agent runs: cost, success rate, latency.
OpenTelemetry: Protocols and ops; An open standard for traces and metrics; MCP documents trace context propagation with it.
Orchestrator-worker: Patterns; A lead agent splits work, workers execute, the lead merges.
Permission modes: Coding agents; Settings for what the agent may do without asking: read only, edit, run commands, bypass.
Plan mode: Coding agents; Claude Code mode where the agent proposes a plan before editing.
Plugin: Coding agents; A bundle of skills, hooks, commands, agents and MCP servers.
PostToolUse: Coding agents; Hook event after a tool runs; used for formatting, linting, logging and feedback.
PreCompact: Coding agents; Hook event that runs before compaction; used to save what matters.
PreToolUse: Coding agents; Hook event before a tool runs; can block, allow or modify the call.
Progressive disclosure: Concepts; Loading a short summary always and detail only when needed.
Prompt caching: Protocols and ops; Provider reuse of an unchanged prompt prefix, making repeat calls cheaper and faster.
Prompt chaining: Patterns; A fixed sequence of model calls where each output feeds the next.
Prompt engineering: Concepts; Choosing the wording of instructions given to a model.
Prompt injection: Concepts; Untrusted content that tries to act as instructions to the agent.
RAG: Patterns; Retrieval-augmented generation: retrieve relevant documents, add them to the prompt, then generate.
Rate limit: Protocols and ops; A provider's cap on requests or tokens per minute, returned as HTTP 429.
Regression eval: Protocols and ops; Rerunning the eval suite after every change to catch quality drops.
Reranking: Patterns; Reordering retrieved results by relevance with a second, more precise model.
Sandbox: Protocols and ops; An isolated environment, such as a container, VM or worktree, for running agents with wide permissions.
Schema validation: Concepts; Checking that structured output contains the required fields and types.
Secrets handling: Protocols and ops; Keeping keys out of prompts, context, logs and commits.
SessionStart: Coding agents; Hook event at session start, resume or after compaction.
Skill: Coding agents; A folder with SKILL.md and scripts; its description stays loaded, its body loads on demand.
Slash command: Coding agents; A saved prompt triggered by name, such as /review.
Statusline: Coding agents; A command whose output Claude Code shows as a status bar: model, context, cost, session.
stdio transport: Protocols and ops; MCP transport where the server runs as a local process speaking over standard input and output.
Stop condition: Concepts; The rule that ends an agent loop.
Stop hook: Coding agents; Hook event when the agent wants to finish; exit 2 forces it to continue.
stop_hook_active: Coding agents; Field telling a Stop hook that Claude is already continuing because of a Stop hook.
Streamable HTTP: Protocols and ops; MCP's standard remote transport.
Structured outputs: Concepts; Responses constrained to a JSON schema.
Subagent: Coding agents; A separate agent with its own context, started by a parent with a brief.
SubagentStop: Coding agents; Hook event when a subagent finishes.
Supply chain risk: Protocols and ops; Risk from third-party plugins, skills and MCP servers that run with your access.
System prompt: Concepts; Standing instructions that frame every turn.
tmux: Protocols and ops; Terminal multiplexer with windows and panes that survive closing the terminal.
Token: Concepts; The unit of text a model reads and writes; drives cost and latency.
Tool: Concepts; A function the model may ask the harness to run, defined by name, description and input schema.
Tool annotations: Protocols and ops; MCP hints describing tool behaviour, such as read-only or destructive.
Tool schema: Concepts; The JSON schema describing a tool's input; the model reads it to form calls.
Tracing: Protocols and ops; Recording every step of a run, including tool calls, tokens, latency and cost.
UserPromptSubmit: Coding agents; Hook event when the user submits a prompt; can add context or block.
Vector store: Protocols and ops; A database for embeddings and similarity search, such as Qdrant or pgvector.
Verifier agent: Patterns; An agent that re-checks claims against independent evidence.
Workflow: Patterns; Fixed steps defined in code, with model calls at some steps.
Worktree: Coding agents; A git feature for multiple checkouts of one repo; one per parallel agent.
Agent Skills: Coding agents; An open standard for packaging procedures as folders with a SKILL.md file plus optional scripts and references; specified at agentskills.io.
Agentic AI Foundation: Protocols and ops; A Linux Foundation body launched in December 2025 to steward MCP, AGENTS.md and goose as open standards.
Builder harness: Concepts; The part of a coding agent's harness shipped by its makers: system prompt, built-in tools, loop, compaction.
User harness: Concepts; The outer harness a team builds around a coding agent for its own codebase.
Guide: Patterns; A feedforward harness control that steers the agent before it acts, such as AGENTS.md or a skill.
Sensor: Patterns; A feedback harness control that checks the agent's work and returns the result so it can self-correct.
Computational control: Patterns; A deterministic guide or sensor executed by code: tests, linters, type checkers, blocking hooks.
Inferential control: Patterns; A guide or sensor where a model does the judging, such as a reviewer agent or LLM judge.
Harness drift: Concepts; Guides and sensors contradicting each other as a harness grows; fixed by auditing the harness like code.
Loop engineering: Concepts; Designing the goal, checks, stop conditions, state and budget so an agent works toward a verified result without a human prompting each step. An emerging term, not yet standardized.
Goal loop: Patterns; Repeating agent runs until a verifiable check passes or a limit is reached.
Scheduled loop: Patterns; An agent run triggered repeatedly by a timer or event.
Verified exit: Patterns; A loop stop condition checked by something other than the agent itself.
No-progress check: Patterns; Stopping a loop when its progress measure stops improving for a set number of rounds.
Structured notes: Patterns; Progress and decisions the agent writes to a file outside the context window and rereads later.
Vibe coding: Concepts; Karpathy's February 2025 term for accepting model-written code without closely reading it.
pass@k: Protocols and ops; Probability that at least one of k trials succeeds; a capability measure.
pass^k: Protocols and ops; Probability that all k trials succeed; a consistency measure.
Trial: Protocols and ops; One attempt at an eval task; agents are run several times per task because results vary.
Grader: Protocols and ops; Logic that scores an agent's output or transcript: code-based, model-based or human.
Transcript: Protocols and ops; The full record of an agent run: messages, tool calls and results.
goose: Coding agents; An open source local agent framework from Block, stewarded by the Agentic AI Foundation.