Agentic Workflow
A workflow template for Claude Code and five more agents: a backlog task, a diff review and a commit.
- 59 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent commits finished work on its own and changes project files; hooks block config writes into the code directory.
- Shared memory stores notes across sessions and hosts.
Install
- Works in
Manual install
git clone https://github.com/varienos/agentic-workflowThe first Quick Start command clones the template; the remaining steps run inside the `Agentbase` directory.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent commits finished work on its own and changes project files; hooks block config writes into the code directory.
- Shared memory stores notes across sessions and hosts.
This is third-party code. Review the repository files before installing.
What it does
The template sets up one development cycle: take a backlog task, implement it, review the diff and commit. Claude Code is only one host; Codex, Gemini, Antigravity, Kimi and OpenCode receive the same steps. The `/bootstrap` command interviews you about the project, `generate.js` writes the shared workflow from the answers and `transform.js` converts it into the chosen host's format. Included are automatic review by three agents (code quality, silent failures, regressions), root-cause bug fixing with at most three hypotheses, the `codebase-guard` and `test-enforcer` hooks, pre-push checks and shared memory through the `basic-memory` MCP server. Tasks live in Backlog.md, agent configuration in `Agentbase/` and project code in `Codebase/`.
Who it is for. Developers and contractors running a project with several agents who want one process for tasks, review and commits.
Good fit when
- Claude Code and Codex both work on a project and the rules must match.
- The agent must take tasks from a backlog, not from random requests.
- A diff needs an independent review before a task closes.
- Agent configuration must be kept apart from the code delivered to a customer.
Not a fit when
- You need one skill or one command, not a whole directory structure.
- You do not want to install Backlog.md and `basic-memory`.
- The agent must not commit by itself.
Example request
Run /bootstrap for my project, then take the first backlog task and take it through review.Limitations
It needs Node.js 18 or newer, Git 2.38 or newer, `jq` for hooks and Backlog.md (`npm i -g backlog.md`). Shared memory needs `uv` and Python 3.12 or newer, and bootstrap installs the `basic-memory` CLI if it is missing. Hooks in `settings.json` run only in Claude Code; other hosts have to replace them with rules. By design the agent commits finished work without secrets on its own and pushes only when asked. Secrets and personal data must never go into persistent memory. Some commands in the repository are examples and are replaced by generated ones after bootstrap.
How to disable. Delete the added `Agentbase/` and `Docbase/` directories, the `basic-memory` entry from the MCP configuration and the hooks from `settings.json`. Remove the `Codebase` link without touching the project itself.
FAQ
Which agents are supported?
Claude Code, Codex, Gemini, Antigravity, Kimi and OpenCode. Output for non-Claude hosts comes from `node transform.js ... --targets codex` and similar.
Why is Backlog.md needed?
The whole task lifecycle runs through its CLI: creation, priority, implementation, review and closure.
What does the automatic review do?
Three agents inspect the diff: code quality, silent failures, regression likelihood. A fourth, devil's advocate view is added for security changes.
Where is memory stored?
In the `Docbase/memory/` vault that every host connects to through the `basic-memory` MCP server.
How do I attach it to an existing project?
Link `Codebase` to the project, run `npm install` in `Agentbase`, run `generate.js` and `transform.js` with the manifest, then `/bootstrap`.
Discussion
No questions or reviews yet.
Related
WorkflowsClaude Code kit with eight phase agents, 21 skills and hooks that keep the agent inside set limits.
Shared Claude Code settings for a team: hooks against dangerous commands, a check before finishing and skills from spec to review
A tool and skill that let Claude Code hand tasks to Claude, Codex or OpenCode subagents in separate Git worktrees
A 33-skill plugin with gates: question the request, verify facts against sources, test first, review by a second model
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
A 13-skill delivery workflow for coding agents: from idea to plan, review, security audit, tests and a shipped pull request
README in short
A process template for several agents: backlog, review, safety hooks and shared memory. It is generated for the project stack from interview answers.