The portable workflow layer for AI companions.

Define multi-step AI workflows in a single file. Your AI reads it and follows the steps.

The Problem

Every AI session starts from scratch. You explain the same process every time: "when reviewing code, check for X, Y, Z" or "when fixing a bug, follow these steps." Your AI has no memory of your preferred workflows.

The Solution

aflow lets you define reusable workflows in a flow.md file. Your AI reads this file and follows multi-step processes consistently.

npx @aman_asmuei/aflow init

This creates ~/.aflow/flow.md with 4 starter workflows. Add it to your AI's system prompt and it follows your workflows automatically.

The Ecosystem

aman
├── acore   →  identity     →  who your AI IS
├── amem    →  memory       →  what your AI KNOWS
├── akit    →  tools        →  what your AI CAN DO
├── aflow   →  workflows    →  HOW your AI works
├── arules  →  guardrails   →  what your AI WON'T do
└── aeval   →  evaluation   →  how GOOD your AI is

| Layer | Package | What it does | |:------|:--------|:-------------| | Identity | acore | Personality, values, relationship memory | | Memory | amem | Automated knowledge storage (MCP) | | Tools | akit | 15 portable AI tools (MCP + manual fallback) | | Workflows | aflow | Reusable AI workflows (code review, bug fix, etc.) | | Guardrails | arules | Safety boundaries and permissions | | Evaluation | aeval | Relationship tracking and session logging | | Unified | aman | One command to set up everything |

Each works independently. aman is the front door.

Quick Start

# Create flow.md with starter workflows
npx @aman_asmuei/aflow init

# List your workflows
npx @aman_asmuei/aflow list

# See a specific workflow
npx @aman_asmuei/aflow show code-review

# Add a custom workflow
npx @aman_asmuei/aflow add deploy

# Remove a workflow
npx @aman_asmuei/aflow remove daily-standup

# Health check
npx @aman_asmuei/aflow doctor

Commands

| Command | What it does | |:--------|:------------| | aflow | First run: create flow.md. After: show workflows | | aflow init | Create ~/.aflow/flow.md with starter workflows | | aflow list | List defined workflows | | aflow show <name> | Show a specific workflow's steps | | aflow add [name] | Add a new workflow interactively | | aflow remove <name> | Remove a workflow | | aflow doctor | Health check |

How It Works

flow.md — The Workflow Definition

Every workflow lives in ~/.aflow/flow.md:

# My Workflows

## code-review
When asked to review code:
1. Analyze for bugs, logic errors, and edge cases
2. Check for security vulnerabilities (OWASP top 10)
3. Evaluate code style and maintainability
4. Summarize findings with severity ratings (critical/warning/info)
5. Suggest specific fixes with code examples

## bug-fix
When asked to fix a bug:
1. Reproduce — understand the expected vs actual behavior
2. Locate — find the root cause in the codebase
3. Fix — implement the minimal change that fixes the issue
4. Test — verify the fix works and doesn't break other things
5. Document — explain what was wrong and why the fix works

Your AI reads this file and automatically follows the defined steps when the trigger matches.

Platform Behavior

| Platform | What happens | |:---------|:------------| | Claude Code / Cursor | AI reads flow.md, executes steps automatically with tools | | ChatGPT / Gemini / Other | AI reads flow.md, guides you through each step manually |

Integration with acore

If you use acore, run acore pull --sync-only to refresh your platform files and pick up flow.md.

Starter Workflows

The aflow init command includes 4 workflows to get started:

| Workflow | Trigger | Steps | |:---------|:--------|:------| | code-review | When asked to review code | 5 steps: bugs, security, style, summary, fixes | | bug-fix | When asked to fix a bug | 5 steps: reproduce, locate, fix, test, document | | feature-build | When asked to build a feature | 5 steps: clarify, design, implement, test, review | | daily-standup | When starting a new session | 4 steps: git log, PRs, summary, focus |

Privacy

All data stays local. ~/.aflow/ contains your workflow definitions. No telemetry. No accounts. No cloud.

Contributing

Contributions welcome! Add starter workflows, improve the parser, or suggest features.

License

MIT

Define once. Follow always. Every AI.