Appendices

Q: What should I trust when a reference page and my tool disagree?

The installed binary. Run claude --help or codex --help for the version you have, because Claude Code ships almost daily and its stable channel runs about a week behind the latest one.

Overview Updated 2026-10-17

The Appendices are the lookup shelf for Claude Code, Codex and Cursor: four reference pages covering commands and flags, configuration files, keyboard shortcuts, and the glossary. Each page states when it was checked. Each page is for finding an exact spelling or file path, which should then be confirmed against the tool version installed on the machine that runs the agent.

This section is for developers who script agents and for tech leads who own a team’s shared configuration. The typical situation: a CI job that worked last month now fails with unexpected argument, or one laptop ignores AGENTS.md while the rest of the team’s machines read it. The answer is a flag, a file path or a release channel, and you need it spelled exactly.

Which reference page answers your question?

You want to know…	Open	What it covers
The exact flag for a headless run, an approval policy or an MCP command	Command reference	CLI flags and slash commands by job, plus a table of removed and renamed flags
Which instruction file a tool reads, and in what order	Configuration reference	`.cursor/rules/`, `CLAUDE.md`, `AGENTS.md`, `.claude/settings.json` and Codex `config.toml`, with loading rules
The key that cancels, accepts or opens a panel	Keyboard shortcuts	Shortcut tables for the Cursor editor and the Claude Code and Codex terminals
What a term means in this documentation	Glossary	The canonical term for each concept, so one word means one thing across the site

How do you check a reference entry against your installed version?

A reference page is correct for the version it names. Claude Code ships roughly one release per day, and its stable channel typically runs about a week behind latest, so two machines on one team can disagree. Treat the installed binary’s help output as the authority.

Print the version you run. Compare it with the version the reference page names.
Terminal window
claude --version claude doctor # installation health; reads settings without a trust prompt
To install the stable build on one machine, run claude install stable; the autoUpdatesChannel setting ("stable" or "latest") keeps it on that channel.
Terminal window
codex --version codex features list # feature flags with their stage and effective state
Check the version in the editor’s About dialog. Cursor facts on these pages are re-checked less often than the two CLIs, so confirm a setting in Cursor’s own documentation before you roll it out to a team.
Check the exact flag in the help for that subcommand. Flags differ between subcommands: codex exec --help lists different options from codex --help. If the flag is missing from the output, it does not exist in your version, whatever any page says.
Pin the version in CI. A flag verified today should still exist on next week’s runner. Install an exact version instead of the moving tag. The example pins the latest build of 2026-10-12 (2.1.296); stable is 2.1.292 on 2026-10-17, so pin that one if your machines follow stable:
Terminal window
```
npm install -g @anthropic-ai/claude-code@2.1.296 @openai/codex@0.162.1
```
Bump the pin in a pull request, so that the CI run on that pull request proves your scripts still run.

Search this repository's shell scripts, Makefiles, package.json scripts and
.github/workflows for every invocation of `claude` and `codex`. For each one,
run the matching help command on this machine (for example `codex exec --help`
or `claude --help`) and report any flag that does not appear in that output.
Check these known traps explicitly: `--full-auto`, `-a on-failure`,
`-a untrusted`, and `codex exec -a` (the approval flag goes before `exec`).
Output a table with file:line, the current command, the problem, and the
corrected command. Do not edit any file.

List every instruction file in this repository: CLAUDE.md, CLAUDE.local.md,
AGENTS.md, AGENTS.override.md and files under .cursor/rules/, excluding
node_modules. For each, state whether it applies to the current working
directory according to the configuration reference rules for the tool you are
running in, and why. Then tell me the exact command or view I should use to
confirm it independently: /context in Claude Code, `codex debug prompt-input`
in Codex, and in Cursor (which has no such command) opening a file the rule
should cover and checking for the rule's canary line, as the configuration
reference describes. Do not rely on your own memory of what you loaded.

The second prompt produces a hypothesis, not proof: a model can misreport its own context. The proof is the tool’s view of the context, which the configuration reference walks through with a canary line per file and a CI budget check. Cursor has no counterpart to /context or codex debug prompt-input on these pages (checked 2026-09-26), so in Cursor the canary line is the only independent check.

When the reference and your tool disagree

Symptom	Likely cause	Recovery
`unexpected argument` or `invalid value` in CI	The flag was removed or renamed, or it sits after the wrong subcommand	Run the first prompt above, then check the removed-flags table in the command reference
A feature the page describes is missing on one machine	That install is on an older pinned version, or on `stable` for a feature from v2.1.293 or later	Run `claude --version` on both machines; align the channel or the pin
A shortcut does nothing in Cursor	The key is rebound in your keybindings, or an extension claims it	Check the keybinding editor for conflicts before assuming the page is wrong
A term means something different in a vendor’s docs	Vendors rename features; the glossary keeps one canonical term	Use the glossary term in team docs and note the vendor’s name beside it
A link you bookmarked for models, pricing or migration no longer lands here	Those pages moved during the 2026-09 restructure	Use the links in the next section

What moved out of the Appendices?

Model versions and prices now live on one page, the models hub, so no other page repeats them. Plan pricing and the cost worksheet are in the pricing analysis, and the tool-switching checklist is in the migration guide. Per-tool release notes sit in each tool’s reference: Claude Code, Codex and Cursor.

Where to go next from the reference

When something fails and the cause is not a flag, go to the troubleshooting guide.
Before you script an unattended run, set the approval and sandbox posture with permissions and sandboxing across the three tools.
To see what changed on this site since your last visit, read what’s new.
For the other help pages, including the FAQ and community resources, open the resources overview.

Frequently asked questions

What is in the Appendices section?

Four lookup pages for Claude Code, Codex and Cursor: a command reference, a configuration reference for instruction and settings files, keyboard shortcuts, and a glossary. They answer "what is the exact spelling", not "how do I work".

Where did the model comparison, pricing calculator and migration checklist go?

Models and prices moved to the models hub in the Ecosystem section. The pricing calculator merged into the pricing analysis and the migration checklist into the migration guide, both under Comparisons.

What should I trust when a reference page and my tool disagree?

The installed binary. Run claude --help or codex --help for the version you have, because Claude Code ships almost daily and its stable channel runs about a week behind the latest one.