Is a CLAUDE.md, AGENTS.md or Cursor rules file enough for your coding agents?
A context file is enough for the standing instructions every task needs: how to build and test, the code style, the folders to leave alone. For the decisions and rules behind each feature, confirmed by the people who know and changing feature by feature, add Rivora: it keeps them as your team designs each feature, and sends the coding agent the approved plan with the decisions and rules attached.
Checked on 10 November 2026 against Anthropic's Claude Code docs, Cursor's docs, OpenAI's Codex docs and agents.md.
What are context files good at?
Each coding agent reads a plain Markdown file of instructions before it starts work. From each vendor's own docs:
- CLAUDE.md (Claude Code) can be set for an organisation, a person, a project or a folder. Claude Code loads it at the start of every session, and
/initdrafts a first version from your codebase. - Cursor rules live in
.cursor/rules. Each rule can apply always, to matching files, when the agent judges it relevant, or only when someone mentions it. Team and Enterprise plans add team rules that admins can enforce. - AGENTS.md is an open format that many coding agents read, including Codex, Cursor and Claude Code. A large repository can keep one per folder, and the nearest file wins.
They sit in the repository, so the whole team shares them through git. For "how we work in this codebase", start here.
Where does a file stop?
A context file holds what someone wrote down and keeps up to date. Each vendor also asks you to keep it short: Anthropic suggests under 200 lines for a CLAUDE.md, Cursor under 500 lines for a rule, and Codex stops reading at 32 KiB by default.
On an established product, the answers a coding agent needs most are feature by feature, and many were never written down:
- a rule that applies to one client only;
- a decision that replaced last quarter's decision;
- the module that's still in the code but no longer used.
A file can't hold every one of those, and nobody reviews each line before an agent reads it. That's the gap between instructions for the codebase and answers for the feature.
What does Rivora add?
- Answers from the people who know. Rivora drafts each part of a feature's solution and raises the questions it can't answer. Your experts answer once.
- Memory the team confirms. Each decision records where it applies, where it came from and who confirmed it. A rule for one client stays with that client.
- Checked before anyone builds. A person approves each part, sends it back with a comment, or edits it. Only approved parts feed the parts that follow.
- Sent to the agent with the plan. Approved work goes to coding agents such as Claude Code, Cursor and Codex, with the decisions and rules attached, and the agent can look up the sources behind each requirement.
The context file tells the agent how to work in your codebase. Rivora tells it what this feature needs, and why.
How do they compare?
| Context file (CLAUDE.md, AGENTS.md, Cursor rules) | Rivora | |
|---|---|---|
| Holds | Standing instructions for the codebase: commands, style, conventions | Decisions, rules and clarifications for each feature, each with its scope and who confirmed it |
| Written by | Whoever edits the file | The people who know, answering Rivora's questions, confirmed by your team |
| Checked by | Code review of the file, if the team does it | A person approves each part |
| Size | Kept short, because it loads into every task | Each feature's approved plan, sent when the agent works on that feature |
| Together | Keeps the agent working your way | Gives the agent what your team confirmed for this feature |
When is a context file the better choice?
For a young codebase, a small team where everyone knows the rules, or any instruction that applies to every task. Most teams using coding agents should keep one. Our free context file builder helps you write a first one.
When should a team add Rivora?
When the agent keeps guessing on an established product: rules for one client, decisions that changed, code that's no longer used. Those answers belong with the feature, confirmed by the people who know.
Questions
- Can't we build this ourselves with Claude skills?
You can build skills that draft the document. The value is the memory behind it, and who checked it: answers from the people who know, confirmed by your team, kept for every feature after.
- Do we still need our CLAUDE.md or AGENTS.md?
Yes. Keep it for how to work in your codebase. Rivora adds what each feature needs.
- Does Rivora change our code or our context files?
No. Rivora reads your code through a GitHub App with read-only access. Your engineers and their coding agents make every change.
Bring one real feature you're planning in the next month.