Replit Agent reads the rules from replit.md
Replit Agent reads replit.md from the project root, not AGENTS.md, so init --agents replit writes the rules to both and rewrites only its own section of replit.md when you run it again. There’s no post-edit hook: Agent runs ultracite fix and check when the rules tell it to, and a Workflow or CI check runs them on demand.
Replit Agent by ReplitChecked against October 2026
# Ultracite Code Standards
This project uses **Ultracite**, a zero-config preset that enforces strict code quality standards through automated formatting and linting.
## Quick Reference
- **Format code**: `npx ultracite fix`
- **Check for issues**: `npx ultracite check`
- **Diagnose setup**: `npx ultracite doctor`
Oxlint + Oxfmt (the underlying engine) provides robust linting and formatting. Most issues are automatically fixable.
…# Ultracite Code Standards
This project uses **Ultracite**, a zero-config preset that enforces strict code quality standards through automated formatting and linting.
## Quick Reference
- **Format code**: `npx ultracite fix`
- **Check for issues**: `npx ultracite check`
- **Diagnose setup**: `npx ultracite doctor`
Oxlint + Oxfmt (the underlying engine) provides robust linting and formatting. Most issues are automatically fixable.
…- 1
Workspace custom instructionsAlways-on guidance for every project in the workspace, set in Workspace Settings. Pro and Enterprise. - 2
replit.mdUltraciteRead automatically when Agent processes a request. It must be in the project root. - 3
AGENTS.mdUltraciteNot in Replit’s docs.initwrites it for your other agents and copies the same rules intoreplit.md. - 4
.agents/skills/Installed skills: Agent sees each one’s name and description and loads the rest when a task needs it.
How Replit Agent reads the rules
Agent’s project instructions come from one file, replit.md in the root, on top of any workspace-wide custom instructions. Agent also writes and updates replit.md itself, so Ultracite’s section shares the file with Agent’s own notes.
- Root only
replit.mdmust sit in the project root. Agent doesn’t look for it in subdirectories.- Agent edits it
- Agent creates
replit.mdin new projects and updates it as it learns.initrewrites only its own section, from# Ultracite Code Standardson. - Size
- No strict limit, but very large files may not be fully processed. Ultracite’s section is about 120 lines.
- Starting over
- Delete
replit.mdand start a new conversation, and Agent writes a fresh one from your project. Runinitagain afterwards to put the rules back. - Workspace-wide
- Custom instructions load on every message in every project of the workspace. Replit warns that strict requirements in them aren’t guaranteed.
- Other tools
replit.mdgives context to Agent conversations only, not to other AI tools, which is whyinitwritesAGENTS.mdas well.
Checks run when the rules ask for them
Replit’s docs describe no hook that runs after Agent edits a file. Agent follows replit.md, which tells it to run npx ultracite fix and npx ultracite check; for a check that doesn’t depend on Agent, add a Workflow or a CI job.
- **Format code**: `npx ultracite fix`
- **Check for issues**: `npx ultracite check`
- **Diagnose setup**: `npx ultracite doctor`- 01Agent edits files in your Replit App with
replit.mdin context. - 02Replit reviews the changed files for insecure patterns and hardcoded secrets, a security check rather than a lint.
- 03Agent runs
npx ultracite fixandnpx ultracite checkwhen it follows the rules, or when you ask it to before it commits. - 04A Workflow that runs
npx ultracite checkgives you the check on demand, and CI covers what reaches GitHub.
Worth knowing about Replit Agent
Ask for the linter before the commit
Replit’s guide to contributing to a GitHub project suggests a follow-up prompt: “Before you commit, run the project’s linter and test suite and show me anything that fails.” Agent then commits to a branch, pushes it and opens the pull request.
Workflows stop at the first failure
A Workflow in sequential mode runs its tasks in order and stops when one fails, so an Execute Shell Command task for
npx ultracite checkahead of your tests makes a one-click gate in the Workflows pane.A security pass on Agent’s changes
As you build, Replit reviews the files Agent changes for common insecure patterns and hardcoded secrets. It’s a lightweight security check, not a lint, so it sits alongside
ultracite checkrather than replacing it.The Ultracite skill
Replit Agent loads skills from the project’s
/.agents/skillsfolder, which is whereinitinstalls the Ultracite skill, so Agent finds it. It sees every installed skill’s name and description and loads the full skill when a task calls for it.
Replit Agent questions, answered
- Does Replit Agent read AGENTS.md?
- Not according to Replit’s docs: Agent reads
replit.mdfrom the project root.init --agents replitwrites the rules toAGENTS.mdfor your other agents and copies them intoreplit.md, after anything already in it. - Will Replit Agent overwrite the rules in replit.md?
- Agent can edit
replit.mdas it learns about your project. If Ultracite’s section changes or disappears, runnpx ultracite@latest init --agents replitagain: it rewrites its own section and leaves Agent’s notes alone. - How do I make Replit Agent run the linter?
- The rules already tell it to run
npx ultracite fixandnpx ultracite check. To be sure, ask in your prompt, for example before it commits, or add a Workflow that runs the check. - Can I set Ultracite’s rules for every Replit project?
- Workspace custom instructions apply to every project, on Pro and Enterprise. Replit says shorter instructions are followed more reliably, so keep the full rules in each project’s
replit.mdand use custom instructions for a one-line reminder at most. - Does Replit Agent support hooks?
- Replit’s docs don’t describe hooks that run after Agent edits a file. A Workflow and a CI check are the places to run
ultracite checkwithout relying on Agent.
Keep going
- AgentLovableBrowser-based app builder where an agent writes full-stack web apps from chat.
- AgentDevinCognition’s autonomous coding agent, in cloud sessions, a CLI and a desktop app.
- AgentJulesGoogle’s asynchronous agent that works on your repo in a cloud VM and opens pull requests.
- DocsAgent rulesWhat
initwrites for every agent, and how to keep it current.
Ship less slop
One command sets up your linter, formatter, editor and agents. Works with Oxlint, Biome and ESLint.