Agent Glossary
A tiny translator for people who do not want to learn developer slang just to use an AI tool.
The words you will keep seeing
- agent = an AI that can take actions on your behalf, not just answer questions — it can read files, run commands, send messages, check sources, and loop through tasks until they are done
- agentic = working in agent mode — taking actions, not just talking
- prompt = what you type to the AI — the instruction, question, or brief you give it
- context = everything the AI can currently see: your messages, any files you shared, earlier instructions in the same conversation
- context window = the limit on how much context the AI can hold at once — when a conversation gets very long, earlier parts start to fall out of view
- token = the chunks the AI reads and writes in (roughly words or parts of words) — token limits are why very long conversations can get wobbly
- hallucination = when the AI confidently says something that is wrong or made up — it is not lying, it is just pattern-matching its way into a wrong answer
- CLI = the command box, terminal, or text window where you type commands instead of clicking buttons
- SDK = a starter toolkit for developers — a set of ready-made pieces so you are not building from scratch
- MCP server = a bridge that lets an AI tool connect to an external app or service and use it as a tool
- runtime = the thing actually running the code right now
- deploy = put it somewhere usable — usually online or on a server
- local = on your own computer, not online
- repo = the project folder tracked in GitHub
- branch = a safe side copy of the project where you can make changes without touching the main version
- secret = sensitive stuff like API keys, passwords, and tokens that should not be pasted into chat or committed to Git
More words that come up
- build = turn the code into something that can run
- debug = find out why it broke
- scaffold = make the starter structure so you are not beginning with a blank page
- workflow = the steps you follow to get something done
- automation = something that does a task for you without you having to trigger it each time
- signal = the useful stuff worth paying attention to
- noise = the stuff you can safely ignore
A note on branches
A branch is not a forestry crime.
It is a copy of the project where you can make changes safely before merging them back into the main version. “Make a branch” means “make a safe working copy.” Nothing is being cut down.
A note on hallucination
AI hallucination sounds alarming. It just means the model made something up with full confidence.
It happens most with: specific names, statistics, citations, dates, and anything that needs to be factually precise. When in doubt — check it independently.
Why this exists
If a guide starts sounding like this:
“Use the MCP server in your repo scaffold to debug the runtime before deploy”
that is a clue the wording has drifted too technical.
This glossary is here so the repo can say the same thing in English.
A note on secrets
If a tool asks for a secret, do not paste it into a random chat window.
Use the project’s supported secret storage, environment variables, or a secret manager. If you already pasted a real secret somewhere unsafe, assume it is exposed and rotate it.