This article examines Trellis, a Git-tracked engineering framework that solves the "amnesia" problem of AI coding assistants such as Claude Code, Cursor, and Codex. It explains Trellis's three-layer architecture (spec, tasks, workspace) stored under .trellis/, its four-phase workflow (Plan, Implement, Verify, Finish), and its session-level task pointer plus JSONL progressive disclosure mechanism. Compared with single-file entries like CLAUDE.md or .cursorrules, Trellis offers team-shared specs, per-task PRDs, persistent memory journals, and 16+ platform adapters. Limitations include setup cost, team coordination overhead, and best-in-class Claude Code support. The piece situates Trellis within the AI coding evolution alongside Planning with Files, Obelisk, and Hermes Agent, arguing that community focus has shifted from raw model capability to engineering-grade harness design.
Trellis Deep Dive: Giving Claude Code, OpenCode, and Cursor a Persistent Project Brain
Summary
This article examines Trellis, a Git-tracked engineering framework that solves the "amnesia" problem of AI coding assistants such as Claude Code, Cursor, and Codex. It explains Trellis's three-layer architecture (spec, tasks, workspace) stored under `.trellis/`, its four-phase workflow (Plan, Implement, Verify, Finish), and its session-level task pointer plus JSONL progressive disclosure mechanism. Compared with single-file entries like `CLAUDE.md` or `.cursorrules`, Trellis offers team-shared specs, per-task PRDs, persistent memory journals, and 16+ platform adapters. Limitations include setup cost, team coordination overhead, and best-in-class Claude Code support. The piece situates Trellis within the AI coding evolution alongside Planning with Files, Obelisk, and Hermes Agent, arguing that community focus has shifted from raw model capability to engineering-grade harness design.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177981535