English | 中文

TAFFISH brings reproducibility back to shell-based bioinformatics commands.

TAFFISH stands for Tools And Flows Framework Intensify SHell. It is a shell-native reproducible execution and delivery layer for bioinformatics command-line tools and lightweight workflows.

TAFFISH does not try to replace shell scripting or existing workflow systems. Instead, it turns tool calls, parameters, container runtimes, metadata, and versioned releases into installable, distributable, composable, and verifiable executable packages that still behave like ordinary shell commands.

• Start Here
• What TAFFISH Provides
• Installation
• How TAFFISH Hub Works
• For Users
• For App Developers
• Paper
• Project Status

• A shell-native reproducible execution layer for command-line bioinformatics, starting from the shell commands users already run.
• A shell-native executable package model that turns tool invocations into versioned, container-resolved, installable, distributable, composable, and verifiable command entities.
• Command-level reproducibility for bioinformatics tools and lightweight flows, complementing workflow systems such as Nextflow and Snakemake from the layer beneath their task commands rather than replacing them.
• The TAFFISH DSL and its taffish compiler, which compile .taf scripts into shell scripts while preserving command-line composability.
• taf, a local package manager, app installer, and command runner for TAFFISH executable packages.
• Open-source Common Lisp implementation under Apache License 2.0, with source-build, contribution, and security documentation in taffish/taffish.
• A layered security model covering release payload verification, source commit checks, container digest/platform metadata, smoke gates, and conservative MCP boundaries.
• Versioned app releases using the version-release form, such as 0.1.0-r1.
• Container-aware execution through Apptainer, Podman, or Docker backends.
• Runtime backend overrides for generic container tags through TAFFISH_CONTAINER_BACKEND.
• Hub indexing so users can discover, update, install, and inspect apps from a static GitHub-hosted index.
• Smoke metadata and Hub-side trust gates for containerized apps, including container digest, platform metadata, and smoke status in the index.
• Flow dependency metadata so taf install can resolve required app versions automatically.
• Private/local project installation through taf install --from for apps that are not yet published to the public Hub.
• taf publish --release, backed by ignored release.md drafts for publish messages and GitHub Release notes.
• Runtime mirror configuration for China/Gitee or internal Git service mirrors.
• Installer-managed shell completion files and Vim syntax files for everyday CLI and .taf editing.
• taffish-mcp, a conservative stdio MCP server for safe AI-client access to TAFFISH project, app, Hub, config, history, resources, prompts, read-only TAF compiler helpers, and safe app/project compile previews. See the TAFFISH MCP Guide and AI client setup guide.

Install the TAFFISH CLI from taffish/taffish. That repository is the canonical place for source code, current installers, supported platforms, release versions, runtime dependencies, container backend notes, network notes, source-build instructions, release verification, and troubleshooting.

Quick user install:

curl -fsSL https://raw.githubusercontent.com/taffish/taffish/main/install/install-taffish.sh | sh -s -- --user

System install for shared servers:

curl -fsSL https://raw.githubusercontent.com/taffish/taffish/main/install/install-taffish.sh | sudo sh -s -- --system

For users in China, the Gitee installer downloads from the Gitee mirror and can initialize the China mirror profile:

curl -fsSL https://gitee.com/taffish-org/taffish/raw/main/install/install-taffish.gitee.sh | sh -s -- --user

China/Gitee system install:

curl -fsSL https://gitee.com/taffish-org/taffish/raw/main/install/install-taffish.gitee.sh | sudo sh -s -- --system

China/Gitee note: the Gitee mirror is provided for environments where GitHub raw content is slow or blocked. Some mirror services may restrict anonymous raw downloads of large files, especially for larger macOS binaries. If the Gitee installer cannot fetch a binary, use the GitHub installer with a working network/proxy, download manually after logging in to the mirror, or check the current notes in the taffish/taffish README.

After taf is available locally, the usual starting point is:

taf doctor
taf update
taf search <keyword>
taf install <app>
taf info <app>
taf list

For a pinned release install, platform-specific requirements, Gitee/macOS notes, offline installation, shell completion, Vim syntax files, and detailed troubleshooting, use the current taffish/taffish README.

TAFFISH 0.8.1 is the current stable patch release in the first open-source 0.8.x series. The source repository includes Build From Source, Contributing, and Security Policy documents. The 0.8.1 release keeps the 0.8.0 public interface stable, fixes release.md placeholder detection in taf publish --release, documents optional [meta] and [upstream] metadata, and includes SHA256SUMS, SHA256SUMS.asc, and TAFFISH-RELEASE-KEY.asc for manual checksum and signature verification.

Persistent mirror configuration can be initialized with:

taf config init --china --force
taf update

Runtime config can change the index URL used by taf update and rewrite canonical GitHub app repository URLs during taf install, which makes Gitee or internal Git service mirrors possible without changing the official index schema.

For installed commands, TAFFISH_CONTAINER_BACKEND=apptainer|podman|docker can force generic container tags at runtime. For private or local apps that are not yet published to the public Hub, use taf install --from <PROJECT-DIR>.

For a guided first run, read the TAFFISH Quick Start.

TAFFISH Hub is currently GitHub-based. It publishes executable package metadata for local taf commands rather than running user jobs on a server.

1. Each app lives in its own repository with a root taffish.toml.
2. App repositories publish release tags such as v0.1.0-r1.
3. App repositories build container images in their own GitHub Actions workflows.
4. taffish-index scans the organization, validates new versions, records dependency, platform, container digest, smoke, trust, and upstream metadata, and writes static JSON index files.
5. Users run taf update, then install and run apps locally through taf.

The public web Hub at taffish.github.io is the human-facing browser for this ecosystem. The index repository is the machine-facing data source.

Use TAFFISH Hub to browse available executable packages and use taf locally to install and run them. For installation, CLI behavior, .taf syntax, container usage, and troubleshooting, read taffish/taffish-docs.

TAFFISH app projects are structured executable-package repositories with taffish.toml, src/main.taf, docs/help.md, target/ build artifacts, versioned release tags, and optional metadata for dependencies, platform constraints, containers, smoke checks, Hub trust status, and upstream software sources.

The official Hub is curated by the taffish organization. At the moment, publishing to the official Hub is limited to organization members. Developers who want to contribute apps can contact the maintainer to discuss joining the organization or publishing strategy.

The TAFFISH project is described in the bioRxiv preprint:

TAFFISH: A lightweight, modular, and containerized workflow framework for reproducible bioinformatics analyses

Authors: Kaiyuan Han, Ting Wang, Shi-Shi Yuan, Cai-Yi Ma, Wei Su, Kejun Deng, Xiaolong Li, Hao Lv, and Hao Lin.

TAFFISH is open source and under active development. The local CLI/compiler implementation is published in taffish/taffish under Apache License 2.0. The current public infrastructure is built around GitHub repositories, GitHub Actions, GitHub Packages, GitHub Pages, and a static package index. A dedicated server-backed Hub may become useful later, but the current design intentionally keeps the publishing and indexing path simple, auditable, and easy to reproduce.