Your coding agent's memory lives in git now

What it does

Trellis treats your repository as the source of truth for AI coding context. It stores specs, task PRDs, workspace journals, and team conventions inside a .trellis/ directory, then auto-injects the relevant pieces into agent sessions across the 14 platforms it supports. The framework runs a four-phase loop—plan, implement, verify, finish—using specialized sub-agents to brainstorm requirements, write code, run lint/type-check/tests, and archive results.

The interesting bit

Instead of monolithic CLAUDE.md or .cursorrules files that grow stale, Trellis scopes memory per task and promotes new learnings back into .trellis/spec/ after each job, so institutional knowledge compounds rather than evaporating when the context window closes. Personal workspace journals stay separate from shared specs, which keeps team-wide standards in git without forcing everyone to share their individual AI session history.

Key highlights

• Persists project memory, conventions, and task state in .trellis/ alongside your code
• Orchestrates planning, implementation, verification, and archival through dedicated sub-agents
• Works across multiple AI coding agents and IDEs, not just Claude Code
• Auto-injects targeted context instead of dumping entire rulebooks into every prompt
• Shared specs version with the repo, making team standards reviewable and portable

Verdict

Grab it if you want structured, repeatable AI coding workflows that survive past the current chat session. Skip it if your agent usage is limited to sporadic one-off questions and you don’t need persistent project memory.