What's new
- Two usage modes, chosen at the start of a session. The agent asks which one applies, in your chat language, and does not start UI work until you answer.
- Mode 1 (During): the rules guide planning and execution while you build. This is the existing v2.0.0 behavior, unchanged: rules applied while writing, closing with the 4-block Delivery Gate.
- Mode 2 (After): audit an already-finished project. The agent produces a numbered findings list (
anti-slop/audit-001-YYYY-MM-DD.md) where each item cites the violated rule (R-XX), a one-line reason, and a priority. - Priorities come from the rule tiers, not from free-form scoring: Hard Gate = HIGH, Purpose-Gate = MEDIUM, Quality Locks = LOW.
- Nothing changes without your approval. You pick the numbers, only those get fixed, everything unmentioned stays untouched, and a follow-up report records what changed.
- Audit files accumulate with rising numbers (audit-001, audit-002, ...), so a project keeps a history of its clean-ups.
- Coming soon, v3.0.0: ANTISLOP as an installable skill/plugin for your agent CLI (Claude Code, Codex, Cursor, etc.). Target: end of Q3 2026.