Glossary
The words AlbertCode uses, in one place.
| Word | Meaning |
|---|---|
| Accept | Approval 2: apply a finished change to your repository. Refused if you edited a file it touches meanwhile |
| Agent (custom) | A role Albert works as, written in Markdown, which can only take tools away. See Extending |
| Albert | AlbertCode's agent: the one you talk to |
| Approval 1, approval 2 | The two points where a governed change waits for you: the plan, then the diff |
| Ask only | The mode where questions are answered and changes are only offered |
| Attachment | A file or picture sent with your next message (up to 500 MB) |
| Boundary | The limits a repository's contract sets; also the button on the repository card |
| Command (custom) | A message you send often, used by name: /review src/cart.py |
| Connector | A ready-made MCP server you add in a couple of answers |
| Contract (repository) | albertcode.contract.toml: what AlbertCode may do in a repository. See The contract |
| Direct | The mode that works in your real files, asking before each write or applying them as they come |
| Evidence | What a task leaves: the plan, the commands and results, the checks, the reviews, the diff and the approvals |
| Evidence ledger | The record of everything AlbertCode did, each entry linked to the one before, so a change is detected |
| Governed | The default mode: plan, approval, isolated build, checks, reviews, approval, apply |
| Isolated copy | The private copy of your repository a governed task works in, so your files stay untouched until you accept |
| Journey | The steps a person takes through a feature, which a browser test walks |
| Language checks | The checker each edited file is run through (pyright, gopls…), so Albert fixes what its edit broke |
| MCP server | An outside tool server Albert can use, over the Model Context Protocol |
| Memory | .albertcode/memory.md: what Albert has learned about a repository |
| Mode | How plain requests are handled: Governed, Direct (ask each write or auto) or Ask only |
| Pace | thorough (independent reviews on every change) or fast (none) |
| Plugin | Commands, agents and MCP servers in one shareable folder. See Plugins |
| Preview | Your app, run with one command, beside the chat |
| Requirement | One thing the app must do, numbered R1, R2… in .albertcode/requirements.json |
| Session | One conversation. Nothing said in one reaches another |
| Skill | A project convention written down in .albertcode/skills/, which changes what the model is told |
| Task | One change: its request, plan, build, checks, reviews and diff |
| Tool (saved) | One of your commands, saved so the model can call it by name |
| Topic | The thing a group of chats is about (a requirement, a test…), shown as one row in the sidebar |
From docs/glossary.md in huynvic/albertcode-swe-agent at a8cef99 (15 Oct 2026), shown unchanged except for links. AlbertCode SWE Agent documentation, CC BY 4.0. Latest on GitHub.