TAUSIK

English

Русский

English

Русский

Appearance

Русский | English

Glossary: what TAUSIK's words mean ​

This page exists because of a count: across the 88 pages marked for the user, gate appears 715 times, slug 467, QG-2 74 — and there was nowhere to look any of them up. The terms below were gathered by counting those pages rather than by imagination: this is what a reader actually trips over.

The order runs from what you meet in the first hour to what you need later.

The work: what a task is made of ​

Task — a unit of work with a goal, acceptance criteria and a journal. Until a task is started the agent may not write code. Everything else hangs off this.

Slug — a task's short hyphenated name, e.g. no-glossary-so-the-vocabulary-is-a-wall. Every command addresses the task by it, and it does not change after creation, because journals and commits already point at it.

Acceptance criteria (AC) — the list of what must be true before the task may close, written BEFORE the work. At least one criterion says how the system must REFUSE; without that only the happy path is ever checked.

Task journal — the record of how the work went and the evidence for closing it. Appended to, never rewritten: evidence you can edit afterwards is not evidence.

Story and epic — two levels of grouping. The epic answers "which problem are we solving", the story "which part of it is being taken now".

Complexity: simple, medium, complex. It decides which gates are hard and which model the framework recommends.

Tier — how much evidence a closure owes, from trivial to deep. Complexity answers "how much work", tier answers "how much proof".

The checks: what refuses, and when ​

Gate — a check that can refuse. There are about fifty: linter, tests, file size, drift between source and the deployed copies. A blocking gate stops the action; a warning one only speaks.

QG-0 — the gate at the START of a task. It refuses work without a goal and acceptance criteria, because a task with no criteria is closed on a feeling.

QG-2 — the gate at the CLOSE. It refuses a closure with no fresh green verification belonging to that task.

Verify-First — the order in which heavy checks run once, ahead of time, via verify and are cached, so the closure only presents the result. Otherwise every close would cost a full test run.

Receipt — a signed record that a check happened: what ran, with what result, by whom. A forged receipt does not pass verification.

Ratchet — a number allowed only to go down. It is how the framework holds a remainder that cannot be cleared in one pass: rotted references, indistinguishable tests, answer length. Growth goes red; shrinking passes.

Dead end — a recorded approach that did NOT work, and why. It exists so the next person does not spend a day on the same thing.

How it is put together ​

Harness — everything the framework deploys into a project around the agent: skills, hooks, MCP servers, the rules file. It is what turns rules into mechanisms.

Deployed profiles — per-editor copies of the harness (.claude/, .cursor/, …). The copy is what runs, so an edit to the source that has not reached it has not taken effect.

Projection — the Markdown tree under tausik/ into which tasks, decisions and memory are serialised. The database does not travel between machines; the tree does, which is how state moves.

Stack — the project's language and tooling. It decides which gates apply.

Handoff — a session's note to the next one: what was done, what is left, what the owner decides.

Snapshot — the public copy of the repository: the filtered tree of a tag, without the internal projection or the working documents.

Standards ​

SENAR — the external standard for AI-agent discipline that TAUSIK claims conformance to. The rules about a task before code, about the journal, and about a check that did not run certifying nothing come from there.

RENAR — the standard for describing requirements: specifications, traceability, acceptance tests. TAUSIK declares itself NON-conformant to it, and the reason is named on the compliance page.

See also ​