These rules fix the spelling of domain terms, distinguish near-synonyms, or ban misleading words. They target naming drift that is easy to miss in review.
Define terms that are easy to confuse and say where the convention applies. Do not change literal identifiers or quoted source to match a prose rule.
The term uses title case in prose. Literal code tokens keep their original spelling, including inside fenced blocks, and anti-pattern quotes may show the wrong form when teaching the rule.
AGENTS.mdL8–10
Source excerpt starting at line 8.
8Write **Dag** (title case) in all prose. Keep the all-caps or lowercase9spelling only when reproducing a literal code token — never rewrite these,10even inside fenced code blocks:
The glossary separates the contributor agent, maintainers, end user, provider, and client. It also defines environment, project, thread, and turn so discussions about this agent product use the same terms.
AGENTS.mdL47–52
Source excerpt starting at line 47.
47- **you** means the agent reading this file and changing T3 Code.48- **we, us, and maintainers** mean Theo, Julius and the people building T3 Code. These are who you are talking to now.49- **user** means the person using T3 Code to direct coding agents.50- **agent** means the coding agent a user runs inside T3 Code. Depending on context, that may also include you.51- **provider** means the agent runtime or harness T3 Code talks to, such as Codex, Claude, Cursor, or OpenCode.52- **client** means the web, desktop, or mobile UI.
Hardware acceleration is allowed to vary by host. The guide distinguishes that from a missing user-visible feature and rejects disabling a working optimization everywhere as a way to satisfy parity.
AGENTS.mdL15–17
Source excerpt starting at line 15.
15- The parity rule covers user-visible BEHAVIOUR. Hardware acceleration varies by host by design (CUDA/MPS/DirectML, Triton availability, `torch.compile`); skipping an optimization where it physically cannot work is not a parity violation.16- Do not "fix" a parity finding by disabling a working optimization everywhere. That trades a real regression for a semantic one.17- A feature the user can see and use on one OS but not another IS a violation. Judge by what the user can do, not by how fast it runs.