Fluent skill
A skill for Codex and Claude Code that turns ideas and bugs into working code through a writer, tester and reviewers
- 125 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Agents write code and run tests autonomously in a sandbox, and landing can happen under a preset policy.
- Remote execution through AWS sends project code to external machines.
Install
- Works in
Manual install
npx skills add mrinalwadhwa/fluent --skill fluentThe README command, run from the project directory; macOS only for now.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents write code and run tests autonomously in a sandbox, and landing can happen under a preset policy.
- Remote execution through AWS sends project code to external machines.
This is third-party code. Review the repository files before installing.
What it does
Fluent is a software factory you drive from a conversation in Codex, Claude Code, Pi or another agent with skills. The skill turns conversation into an interface and runs Fluent's command-line machinery. Ideas, bugs, user feedback, production logs and agent traces are recorded as observations that grow into work items. Each attempt runs in its own git worktree: a writer produces the code, parallel reviewers and one independent tester check the result, and a learner role saves lessons into project expertise. Work can run in a local sandbox or on a remote machine through AWS Fargate. Landing happens after your authorization or under a policy you set in advance.
Who it is for. Teams and solo developers ready to trust agents with the chain from idea to merge who want to keep verifiable behaviour specifications.
Good fit when
- An idea is only roughly known and has to become a brief, behaviour specifications and a plan.
- The task list grows and part of the work can be delegated to agents through a queue.
- Every commit must pass independent review and a separate tester.
- Lessons from past edits need to accumulate as expertise and not vanish between sessions.
Not a fit when
- You need Linux or Windows: the skill's installation is macOS only for now.
- The change is small and urgent: the brief, specification, plan and review chain takes longer than a manual edit.
- Policy forbids autonomous agents with write access to a repository.
Example request
Let us shape this idea: users want report export to CSV. Ask one question at a time and draft a brief and behaviour specifications.Limitations
Installing through `npx skills add` currently works on macOS only. Autonomous Codex work needs a standalone executable or an official npm installation with a matching platform package: an incomplete installation is rejected before work starts. Remote execution currently supports AWS Fargate. The factory uses its own `fluent` binary, updated with `fluent update` and a checksum check. Older work items from before the split-record format convert explicitly with `fluent work-item upgrade`.
How to disable. Delete the skill folder from `.agents/skills/fluent`, `.claude/skills/fluent` or `.codex/skills/fluent` and the `skills-lock.json` entry if it remains. The `fluent` binary and the project's `.fluent` directory are removed by hand.
FAQ
How do I invoke Fluent after installing?
Tell the agent you want to use Fluent, or invoke the skill explicitly: `$fluent` in Codex, `/fluent` in Claude Code, `/skill:fluent` in Pi.
What does the first command install?
A small bootstrap. On the first invocation `fluent skills add` replaces it with the full skill and removes Fluent's entry from `skills-lock.json`.
Who writes, reviews and tests the code?
A writer implements the change and tests. Parallel reviewers write reports returning pass, fail or uncertain, and one independent tester runs the final check.
Where does the work happen?
In an isolated git worktree separate from your working copy. Execution can use a local sandbox or a remote machine through AWS Fargate.
How do I upgrade Fluent?
With `fluent update`: the binary downloads with a checksum check and existing skill installations refresh.
Discussion
No questions or reviews yet.
Related
SkillsA Claude Code skill that drives Codex over JSON-RPC: tasks, code reviews, parallel research and messaging between two agents.
Skill and plugin for Codex and Claude Code: the agent drives a project through a plan-to-maintain loop with human gates.
lencx skills
lencx-skills
Two skills for Claude Code and Codex: keel works through architecture questions, coding-protocol keeps code edits inside the task
A Claude Code skill that delegates tasks to OpenAI's Codex CLI for code analysis, review and edits
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
A skill for second-agent review of plans or code: Claude Code starts rounds, Codex reviews, and they exchange files
README in short
Two queues are described: one waits for people with the right context, the other waits for agent compute. Landing and deployment watching happen by authorization or policy.