agent-feed: Govern AI Coding Workflows in Your Repository

Overview

Agent Feed is a Python CLI that installs workflow instructions and supporting checks into a code repository. It addresses the fragmentation that occurs when AI coding guidance lives across chat history, tool-specific rules, and scattered project documents.

The installed protocol uses AGENTS.md as a shared entry point, with rules, skills, project knowledge, verification steps, and adapters for Codex, Claude Code, and Cursor. It is designed for teams that want AI-assisted changes to follow repository-defined scope, decision, review, and handoff practices.

Key Features

• Initializes a repository with AGENTS.md and a structured .agents/ workflow.
• Provides thin adapters for Codex, Claude Code, and Cursor, keeping the shared protocol canonical.
• Defines task boundaries, decision checkpoints, verification gates, review, and session handoff.
• Checks protocol structure, references, scripts, skills, and client adapters.
• Supports project-specific constraints and domain knowledge alongside reusable workflow rules.
• Manages local and imported skills, with trust checks for external assets.
• Includes status, upgrade, adapter sync, and configuration commands in its CLI.

Use Cases

• A team using several AI coding tools wants repository instructions to stay consistent across clients.
• Developers want small coding requests to remain scoped and architecture or contract decisions to receive human confirmation.
• Teams need completion to include project-defined tests, documentation checks, or review rather than relying on chat claims.
• A developer working across long AI sessions wants a concise handoff of active conclusions instead of replaying conversation history.
• Maintainers want to add team-specific skills and constraints without replacing a shared workflow foundation.

Getting Started

Install with one of the supported package managers, then initialize and check the repository:

uv tool install agent-feed
agent-feed init
agent-feed check

See the README and Usage Guide for setup options and customization.

Alternatives

• compound-engineering-plugin: It packages reusable skills for planning through implementation and captured learnings, rather than installing a repository-owned workflow with shared checks and gates.
• agent-skills: It offers portable engineering workflows for planning, testing, review, and release, while agent-feed centers on a repository-owned setup with tool adapters.
• agent-skills: It provides reusable specification, review, testing, and sprint skills, rather than a repository-owned workflow with shared instructions and checks.
• agentic-engineering: It emphasizes documentation, session memory, and human-controlled merges, while agent-feed installs shared workflow gates and tool adapters in a repository.

Considerations

• The measured project status is quiet: there have been no commits in the recent measured period, despite releases during the past year. Release activity does not by itself establish current maintenance.
• The project has one contributor, with nearly all contribution activity concentrated there. This creates a maintainer-continuity risk for a workflow teams may depend on.
• The CLI requires Python 3.11 or newer. The npm package is a wrapper and also requires Python on the machine.
• Agent Feed installs and validates instructions and workflow assets; it is not a replacement for the coding agents themselves or for project-specific test infrastructure.