Glossary
Plain meanings for the words people use around AI coding agents, and around Syndai. Written for people who do the work.
A
- Agent babysitting
- The unplanned work of watching, steering, and rescuing a coding agent mid-task, instead of handing off a task and reviewing a finished result.
- Agent memory
- Knowledge an AI agent keeps across sessions, project decisions, house rules, past mistakes, so each run does not start from zero.
- Agent-ready repository
- A codebase set up so a coding agent can find its way, build, test, and contribute without a human guiding every step.
B
- Background coding agent
- A coding agent that runs a task on its own, on managed machines, and returns a result you can review, usually a pull request.
C
- Coding agent harness
- The tools, workspace, limits, checks, and feedback loops around a coding agent's model that turn raw skill into steady work.
- Context engineering
- The craft of choosing what enters a model's context window, files, rules, docs, history, so it reasons about your real system.
P
- Production-ready PR
- A pull request that arrives built, tested, and tightly scoped, so it can be reviewed and merged without a round of rework.
R
- Review council
- A code-review pattern where several reviewer agents check a change from different angles before it is accepted.
- Run receipt
- A record shipped with an agent's delivery that shows what really ran: the commands, the checks, the test output, and the results.
S
- Sandbox lane
- A walled-off, throwaway workspace where an agent's run gets its own dependencies, services, and limits, and cannot touch anything else.
- Spec-driven development
- A way of working where the coding agent builds against a written, reviewable spec instead of a live chat.
V
- Verification-first development
- An approach where the checks that would accept a change are set up first, so the checks decide when the work is done.