| Coding agent |
An AI tool that carries out a job described in natural language on its own by reading files, editing code and running commands |
| Definition of done |
Conditions written so anyone can check in the same way whether a task is finished |
| User flow |
The sequence of steps users take from entering to completing their goal |
| Vertical slicing |
Dividing a large feature into results users can actually use |
| plan mode |
Claude Code's permission mode for reading and exploration without editing source files |
| Repository |
A folder that holds a project's files and change history |
| Instruction file |
A shared team memo the agent reads when starting a task (AGENTS.md, CLAUDE.md, GEMINI.md) |
| Skill |
A file containing procedures or reference knowledge for use as needed. By default, its description loads first and its body loads when used |
| Hook |
A mechanism automatically executing an action such as a command when a defined event and condition match |
| Context |
Everything the agent refers to within one conversation: messages, files read and command output |
| Running locally |
Running an app on your computer before putting it on the internet |
| Test |
Code automatically checking whether specified inputs produce expected results |
| Type check |
A check that value types and function-call forms are correct |
| Linting |
A check finding code problems such as unused variables or risky patterns |
| Build |
The process of packaging source code into a deployable form |
| Stack trace |
A record listing function calls leading to an error, with files and line numbers |
| Exit code |
A number returned when a command ends. Zero means success; other values mean failure |
| Verification loop |
Repeating edits, checks and reading results until checks pass |
| CI |
Continuous integration: a system automatically running checks for each change |
| Status code |
A number indicating a server response's outcome. 200s mean success, 400s client errors and 500s server errors |
| Flaky test |
A test alternating between passing and failing without code changes |
| Git |
A distributed version control tool that records file changes and lets you undo them |
| Commit |
A save point confirming and recording changes |
| Staging area |
Where changes selected for the next commit are gathered (the index) |
| diff |
A line-by-line view of the difference between two states |
| Branch |
A separate line of work diverging from the main line |
| Pull request (PR) |
A request to review branch changes before merging them |
| Merge conflict |
A state where two changes edit the same place differently and Git cannot combine them automatically |
| revert |
A Git command creating a new commit that undoes an existing commit's effect |
| reflog |
A record of branch-tip movements, used to find lost commits |
| bisect |
A Git feature finding a problematic commit by repeatedly halving the range between good and bad states |
| Worktree |
A separate working folder attached to one repository, allowing several branches to be checked out at once |
| Checkpoint |
An agent's saved file state from before editing, enabling rewind. Does not replace Git |
| Secret |
A value granting authority to its holder, such as an API key or password |
| Credential |
A token, key or password proving authority to access a system |
| Environment variable |
A value passed to an app by its execution environment from outside the code |
| Push protection |
A GitHub feature blocking a push before it reaches a repository when a supported secret format is detected |
| Principle of least privilege |
Granting only the permissions needed for a task |
| Permission mode |
A setting defining what an agent may do without asking |
| Sandbox |
An isolation mechanism restricting running commands' file and network access at OS level |
| Prompt injection |
An attack in which instructions hidden in input or external material change the model's behavior in an unintended way |
| Dependency |
An external package used by your app |
| Supply-chain attack |
An attack introducing malicious code through the chain of dependencies |
| Typosquatting |
Publishing malicious packages with names spelled similarly to popular packages |
| Lock file |
A file recording exact installed dependency versions |
| Deployment |
Making a successful build available on the internet. Vercel creates a unique URL per deployment |
| Preview |
A deployment environment for checking changes before production |
| Production |
The live environment used by real users |
| Smoke test |
A quick check of core flows immediately after deployment |
| Monitoring |
Continually observing metrics such as error rates and response times |
| Rollback |
Replacing a problematic deployment with an earlier working one |
| Expand–contract |
Splitting data structure changes across deployments to keep rollback safe at any point |
| Feature flag |
A switch enabling or disabling a feature independently of deployment |
| Version skew |
Errors arising when client and server run different deployment versions |