Terms about proving a change
Unit test
A small piece of code that checks one piece of your software does what it should. It is run automatically and either passes or fails. For an AI agent, a unit test is how a fix becomes evidence: the agent cannot argue with a test.
Failing test
A test that does not pass. It sounds like a bad thing and is the most useful object in bug fixing. A test that fails because of a bug, and passes once the bug is fixed, proves the fix. Bugstack writes one before it writes any fix.
Regression test
A test kept so that a bug that was fixed cannot quietly come back. Every fix Bugstack opens carries the test that reproduced the bug, and that test stays in your repository as a regression test.
Flaky test
A test that sometimes passes and sometimes fails on the same code. Flaky tests teach a team to ignore red builds. Bugstack reports a check that fails and then passes on one commit as flaky, and does not write a fix for it.
Test suite
All of a project’s tests together. A fix has to leave the whole suite passing, not only the new test.
Sandbox
An isolated place where code can run without reaching your secrets, your network or your live systems. See what a sandbox is for.
Terms about getting a change in
Repository
The place your code and its full history are kept. Bugstack works with repositories on GitHub.
Pull request
A proposed change to your code, shown as a list of what would be added and removed, which someone accepts or rejects. Every change Bugstack makes arrives as a pull request.
Merge
Accepting a pull request so its change becomes part of your code.
Revert
A change that undoes an earlier change. When a fix Bugstack merged turns out wrong, undo opens a pull request that reverts it.
CI (continuous integration)
A service that runs your tests and checks automatically on every proposed change. “CI is green” means everything passed. “CI is red” means something failed.
Default branch
The main line of your code, usually called main. A failing build here blocks everyone, which is why Bugstack starts work when it goes red.
Deploy
Putting a new version of your software live. A failed deploy leaves the old version running.
Terms about packages
Dependency
A piece of software written by someone else that yours is built on. Most apps have dozens. Also called a package.
Lockfile
A file that records the exact version of every package your app uses, so every install is identical. An upgrade has to update it or installs fail. Bugstack rebuilds the lockfile in every upgrade pull request.
Major version
A release that may break things that used to work, shown by the first number changing: 2.4 to 3.0. Bugstack holds major upgrades for a person by default.
Vulnerability advisory
A public notice that a version of a package has a security flaw, and which version fixes it. Bugstack checks your packages against these every four hours.
Secret
A password or key that lets your app use another service. It should never be in your repository. See secret scanning.
Terms about agents
AI agent
An AI system that works in a loop: it acts, checks the result and decides what to do next, using tools such as a file editor and a test runner. See agentic coding.
Autonomy level
A setting for how far an agent may go without a person. Bugstack has three: Suggest, Review and Autopilot. See autonomy levels.
Proof setting
On Bugstack, what has to be true before a change merges on Autopilot. The default is a sandbox proof and green CI.
Human in the loop
A person approves each action before it takes effect. See in the loop or on the loop.
Terms about the category
Software maintenance
The work that keeps software running after it is built. See the four types.
Autonomous software maintenance
Software upkeep done by AI agents that find the work themselves and hand it back as pull requests. See the full definition.
Technical debt
Work skipped to ship sooner, paid for later in slower changes. See how it gets paid down.
Questions
What is the difference between a unit test and a failing test?
A unit test is a kind of test. A failing test is any test that currently does not pass. The valuable one is a unit test that fails because of a specific bug, since it passes only when that bug is fixed.
What does “CI is green” mean?
Every automatic check on a proposed change passed. It tells you the change broke nothing your tests cover.
Why do AI agents need a sandbox?
Because they run code nobody has reviewed. A sandbox keeps that code away from your secrets and systems, and makes a passing test mean the code passed.