DISCOVER THE FUTURE OF AI AGENTS

Trellis

Added Apr 24, 2026
Agent & Tooling
Open Source
Workflow AutomationMulti-Agent SystemAI AgentsAgent FrameworkCLIAgent & ToolingDeveloper Tools & CodingAutomation, Workflow & RPA

A multi-platform AI coding workflow orchestration framework providing a unified collaboration substrate across 14 AI coding tools through auto-injected specs, structured task management, and parallel agent execution.

Trellis is a multi-platform AI coding framework developed by Mindfold, positioned as "The best agent harness." Its core approach establishes a structured context management layer above AI coding tools, addressing issues of inconsistent conventions, context fragmentation, and parallel conflicts when teams use multiple AI coding tools.

The framework revolves around the .trellis/ directory convention: spec/ stores project specifications with auto-injection into every session; tasks/ manages PRDs and task lifecycles; workspace/ enables cross-session memory through journals; scripts/ hosts Python automation scripts. Eight built-in slash commands cover the full workflow from requirement discussion (/brainstorm), pre-coding spec checks (/before-dev), compliance validation (/check), to pre-commit checklists (/finish-work).

Parallel execution is built on git worktrees, where the /parallel command launches multiple AI agents simultaneously in isolated branches. For multi-platform adaptation, Trellis supports 14 platforms including Claude Code, Cursor, and GitHub Copilot, auto-generating platform-specific config files on initialization while maintaining a shared .agents/skills/ layer.

The project uses a pnpm monorepo architecture with the CLI written in TypeScript (Vitest-tested), automation scripts in Python, distributed via npm (@mindfoldhq/trellis), currently at v0.4.0 under AGPL-3.0. Requires Node.js ≥ 18 and Python ≥ 3.10, with Windows support since v0.3.0.

Key Use Cases#

  1. Convention persistence: Write database naming rules into .trellis/spec/backend/database-guidelines.md for auto-injection across all sessions.
  2. Parallel feature development: Use /parallel to launch multiple agents, each working in isolated worktrees with self-checks and draft PRs.
  3. Cross-session context continuity: /record-session saves summaries; hooks auto-load previous deliverables/issues/todos on next startup.
  4. Multi-tool team unification: trellis init --cursor --claude unifies spec and task reading across different tools.

Version Evolution#

VersionKey Changes
v0.4.0Command consolidation (before-dev, check), new /update-spec, Python script refactor
v0.3.6Task lifecycle hooks, custom template registry, parent-child subtasks
v0.3.4Qoder platform support
v0.3.0Platform expansion from 2 to 10, Windows compatibility, remote spec templates

Related Projects

View All

STAY UPDATED

Get the latest AI tools and trends delivered straight to your inbox. No spam, just intelligence.