Trellis CLI

A workflow layer for AI-assisted coding: specs, tasks and session journals live in the repository and are injected into the agent's context

  • 15,003 ★ on the repository
  • Last commit
  • AGPL-3.0
CLI

Broad access

Why this level

  • python3 hooks run in every agent session, and `trellis channel` starts worker agents as separate processes
  • The scripts write journals and the task archive into the repository and commit them to git themselves
  • Spec templates with the `-t` flag are downloaded from GitHub at the registry address
All reasons and checks

mindfold-ai/trellis

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add cli/trellis

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • python3 hooks run in every agent session, and `trellis channel` starts worker agents as separate processes
  • The scripts write journals and the task archive into the repository and commit them to git themselves
  • Spec templates with the `-t` flag are downloaded from GitHub at the registry address

This is third-party code. Review the repository files before installing.

What it does

Trellis installs from npm with `npm install -g @mindfoldhq/trellis@latest`; then `trellis init` in the repository root creates a `.trellis/` folder and files for the agents you pick. `.trellis/spec/` holds project rules, `.trellis/tasks/` holds tasks with their requirements and the context for implementation and review, and `.trellis/workspace/` holds each developer's personal session journals. For Claude Code, Cursor, Codex, OpenCode and 18 more platforms the command adds commands, skills, subagents and hooks, and flags such as `--cursor` and `--codex` narrow the list. Work runs in four phases. The `trellis-brainstorm` skill asks one question at a time until the requirements are clear and writes `prd.md`; it does not move on to implementation until the user has replied at least once. The `trellis-implement` subagent writes the code, taking the relevant rules from `implement.jsonl` and `check.jsonl`. The `trellis-check` subagent compares the changes with the specs, runs lint, type checks and tests, and fixes what it finds. `/trellis:finish-work` archives the task and appends to the journal, and `trellis-update-spec` moves new lessons back into `.trellis/spec/`. In Claude Code, python3 hooks fire on session start, clear and compact and before a subagent launches, to inject context. Utility commands sit alongside: `trellis mem` searches conversation history across Claude Code, Codex, OpenCode and Pi, `trellis channel` runs worker agents as separate local processes with an event log, and `trellis update` refreshes the files after a new release.

Who it is for. For developers and teams who use several AI agents in one repository and want rules, tasks and session memory stored next to the code.

Good fit when

  • The agent forgets the project's conventions in every new session and you keep repeating them by hand
  • The team wants coding rules that live in the repository and go through review like any other code
  • A large task has to be written up as requirements first and only then handed to implementation and review
  • Claude Code, Cursor and Codex all work in one project and need a shared file structure
  • You need a record of past sessions: what was done last time and which decisions were made

Not a fit when

  • A single rules file is enough: `CLAUDE.md` and `AGENTS.md` cover that, while Trellis adds a directory of tasks, specs and scripts
  • The environment has no Python 3.9 or newer: the hooks and task scripts are written in Python
  • You need a license softer than AGPL-3.0: read its terms before bringing the tool into a commercial project
  • You only want a proposal, requirements and tasks loop without journals or subagents: OpenSpec in the catalog is closer

Example request

I want to add PDF report export to the project. Start with brainstorming, ask questions one at a time and move to code only after I agree

Limitations

The license is AGPL-3.0. Node.js 18 or newer and Python 3.9 or newer are required.

  • The scripts commit journals and the task archive to git on their own. To turn this off, set `session_auto_commit: false` in `.trellis/config.yaml`.
  • The `.trellis` folder can hold sensitive task and spec text. `trellis ablate` stores a copy of its contents outside the project and keeps it until `trellis restore` succeeds.
  • If a file managed by Trellis changes while the project is ablated, `trellis restore` refuses to write until you resolve the conflict.
  • The set of files depends on the platform: Copilot gets hooks, Pi and Oh My Pi get extensions, the rest get commands or skills. `trellis platforms` prints the full list.
  • The version is 0.x: file formats change through migrations, and `trellis update --migrate` applies them.

How to disable. The `TRELLIS_HOOKS=0` variable disables only the hooks. `trellis ablate` temporarily removes all Trellis files from the project for a comparison, and `trellis restore` brings the earlier state back. `trellis uninstall` deletes the platform files and `.trellis/` for good; run it with `--dry-run` first to see the list.

FAQ

How is Trellis different from CLAUDE.md, AGENTS.md and .cursorrules?

Those files are entry points and tend to grow into one large document. Trellis adds scoped specs, task descriptions, checks between stages, session journals and files for each platform around them.

Do I have to write every spec by hand?

No. The `trellis-spec-bootstrap` skill helps draft specs from existing code, and the important rules are worth tightening by hand. A ready-made template comes with the `-t` flag, for example `trellis init -t electron-fullstack`.

Where does `trellis init -t` get its templates?

From the mindfold-ai/marketplace registry on GitHub: it downloads the `index.json` list first and then the chosen template. The `-r` flag points to your own registry, for example `gh:myorg/myrepo/specs`.

How does a team avoid conflicts?

Personal journals sit separately for each developer, while shared specs and tasks stay in the repository and go through review like any other file.

How do I compare the project with and without Trellis?

Run `trellis ablate`: it makes a verified recovery copy outside the project and removes the Trellis files. Open a fresh agent session, then run `trellis restore`. The `--dry-run` flag previews the changes.

What does trellis mem do?

It searches conversation history with agents. The subcommands are `list`, `search`, `context`, `extract` and `projects`; Claude Code, Codex, OpenCode and Pi are supported.

Discussion

No questions or reviews yet.

CLI tools

A Rust terminal task board: every task runs in its own git worktree and tmux window while Claude Code, Codex, Gemini CLI and other agents work in parallel

CLIBroad access1.7KRepository stars

A team of AI agents from one request: it plans the task, writes code in parallel branches, tests, reviews and opens a pull request

WorkflowBroad access1KRepository stars

Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper

PluginBroad access2.7KRepository stars

A local dashboard for many Claude Code sessions: each job gets its own context and todo, agents spawn child agents and message each other

CLIBroad access1.4KRepository stars

A terminal app for running several Claude Code, Codex, Gemini CLI and Aider sessions in parallel, each in its own tmux session and git worktree

CLIBroad access8.6KRepository stars

A terminal session manager for Claude Code, Codex, Gemini CLI, OpenCode and other agents: one list with statuses, git worktrees and control from a phone

CLIBroad access1.1KRepository stars

Comparison with similar tools

Entries that share the main search query, and how they differ.

EntryOfficialRepository starsMain difference
Trellis (this entry)no15,003
Spec Kitno142,350the specify CLI installs spec-driven development commands and skills, without session journals or history search
OpenSpecno72,697a proposal, requirements, design, tasks, code loop, without session memory
Superpowersno303,510a set of development-process skills without its own file structure or journals in the repository

README in short

The description comes in English and Chinese. It opens with a capability table, followed by the requirements, three quick-start commands and four working steps. After that come a diagram of the four-phase loop, a table of links to the docs at docs.trytrellis.app and a question section covering how it differs from rules files and how to turn Trellis off temporarily. Beside the package the repository holds Trellis's own project with its own specs and tasks, a vitest test suite and a log of migrations by version.

README badge

Report a problem

Foxx AITrellis
Foxx AI

I am Foxx AI and I have already vetted this tool. Ask about install, setup or anything else, and I will keep it simple.