Claude Code Commands & Prompt Reference
Slash commands, prompt input prefixes and hook events of Claude Code, checked against the official commands, interactive-mode and hooks references as published for v2.1.296 (October 2026). Arguments keep the vendor's notation: <path> is required, [name] optional. Rows marked bundled skill are prompts handed to Claude, not coded commands. A command runs only at the start of a message, and some depend on plan or platform. Keyboard shortcuts and CLI flags are on the Claude Code shortcuts page.
Also see: Claude Code keyboard shortcuts & CLI flags.
All Commands
Built-in Slash Commands (119)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /add-dir <path> | Add a working directory | Partial paths get suggestions and Tab accepts one; most .claude/ configuration in that folder is not picked up. |
| /advisor [model|off] | Toggle the advisor model | Accepts fable, opus or sonnet, or any full model ID; run it bare for a picker. |
| /agents | Subagent setup reminder | Now only prints a hint to ask Claude or edit .claude/agents/; v2.1.197 and earlier opened a management interface. |
| /artifact-capabilities | Artifact capabilities reference | Bundled skill listing what a published page can do on your account, such as calling connectors or offering downloads. |
| /artifact-diagramming | Artifact diagram guidance | Bundled skill on when a diagram helps and how to keep inline SVG readable in light and dark themes. |
| /artifacts | Browse your artifacts | Pick an owned or shared artifact, then attach it here, open it in a browser or copy the link. |
| /auto-mode-setup | Draft auto mode environment | Builds autoMode.environment entries from the project and recent sessions for review; needs Pro, Max or Team and v2.1.228+. |
| /autocompact [auto|<tokens>] | Auto-compact threshold | Takes a size like 500k, or auto for the model's tuned default; the value is saved to user settings. |
| /autofix-pr [prompt] | Auto-fix the branch's PR | Runs in the cloud, fixing CI failures and review comments on the open PR; a prompt changes its instructions, gh required. |
| /background [prompt] | Send session to background | Frees the terminal while the session keeps running as an agent; claude agents monitors it, and /bg is an alias. |
| /batch <instruction> | Parallel codebase-wide change | Bundled skill: once you approve its plan, 5 to 30 subagents each build one unit in their own worktree. |
| /branch [name] | Branch the conversation | Switches you into the new branch and keeps the original for /resume, unlike /fork, which runs a background copy. |
| /btw [question] | Ask a side question | The answer never enters the conversation history; run it bare to reread your latest side question. |
| /bug [report] | Report a bug | A consent screen sets how much history is sent; on third-party providers the report is saved locally. Alias /share. |
| /cd <path> | Change working directory | Moves the session to another folder without losing the conversation; Tab accepts a suggested path. |
| /chrome | Claude in Chrome settings | |
| /claude-api [migrate|upgrade|managed-agents-onboard|prompt-audit|cost-optimize|build-eval|hillclimb|preserved-thinking-migration] | Claude API reference skill | Bundled skill that also loads itself when your code uses an Anthropic SDK (anthropic or @anthropic-ai/sdk); subcommands like migrate run one task. |
| /claude-in-chrome [task] | Run a browser task | Bundled skill that drives Chrome to test pages, fill forms or read console logs; needs Chrome integration, e.g. claude --chrome. |
| /clear [name] | Start a new conversation | A name labels the old conversation in the /resume picker; aliases /reset and /new, while /compact keeps the thread. |
| /code-review [low|medium|high|xhigh|max|ultra] [--fix] [--comment] [--max-findings n|all|default] [pr#|branch|path] | Review the diff for bugs | Bundled skill: --fix applies findings, --comment posts them to the PR or merge request, ultra runs a deep cloud review. |
| /color [color|default] | Prompt bar color | Eight colors from red to cyan are offered; default resets, and running it bare picks a random one. |
| /compact [instructions] | Summarize to free context | Optional instructions tell the summary what to focus on; the conversation carries on afterwards, unlike /clear. |
| /config [key=value ...] | Open Settings | Pairs such as theme=dark change a setting without the panel, also under -p; /config --help lists keys. Alias /settings. |
| /context [all] | Show context usage grid | Flags context-heavy tools and memory bloat; in fullscreen the per-item breakdown stays collapsed until you pass all. |
| /copy [N] | Copy the last response | N picks an older reply (2 is second-to-last); with code blocks a picker opens, where w writes to a file. |
| /cost | Alias of /usage | Folded into /usage and kept as a typing shortcut that opens the matching tab of that screen. |
| /dataviz [request] | Chart design guidance | Bundled skill: picks chart forms, colors by role and checks the palette for colorblind safety with a bundled script. |
| /debug [description] | Debug the session | Bundled skill; logging starts when you run it unless you launched with claude --debug, and a description narrows the analysis. |
| /deep-research <question> | Cited web research report | Bundled workflow that fans out across many subagents in the background; Claude starts it only when you invoke it. |
| /design [brief] | Draft UI mockups | Bundled skill that publishes artboards as a Claude Design artifact you edit in a browser and export as PNG or PDF. |
| /design-login | Authorize design-system access | |
| /design-sync [hint] | Upload your design system | Bundled skill converting your repo's React components for Claude Design; a first sync checks each component and can take hours. |
| /desktop | Continue in the Desktop app | Needs macOS or x64 Windows plus a Claude subscription; alias /app. |
| /diff | Review working-tree changes | Shows Claude's edits so far next to your other uncommitted work, in a side panel or a dialog depending on the renderer. |
| /doctor [prompt-audit [path]] | Setup checkup | Bundled skill that finds install, settings and CLAUDE.md problems and fixes them once you confirm; prompt-audit audits instructions. Alias /checkup. |
| /effort [level|auto|status|ultracode [on|off]] | Set effort level | Levels run low to xhigh, plus a session-only max; status prints the level, and ultracode on or off toggles ultracode. |
| /exit | Exit the CLI | In an attached background session it only detaches, leaving the session running; alias /quit. |
| /export [filename] | Export conversation as text | With a filename it writes straight to that file; without one a dialog offers clipboard or file. |
| /fast [on|off] | Toggle fast mode | Switching mid-turn takes effect at once, though the running turn keeps its original speed; needs v2.1.205+. |
| /feedback [report] | Send product feedback | Same dialog and consent step as /bug; with Claude-drafted feedback on, running it bare lists the queued drafts to review. |
| /fewer-permission-prompts | Build a read-only allowlist | Bundled skill: ranks the read-only Bash and MCP calls your transcripts show most often and allowlists them in project settings. |
| /focus | Toggle focus view | Hides everything but your latest prompt, a tool-call summary line and Claude's final reply; needs fullscreen rendering and persists. |
| /fork [prompt] | Fork into a background session | You keep working here while the copy runs; with a prompt it starts at once, otherwise it waits in agent view. |
| /goal [condition|clear] | Set a completion goal | Work continues turn after turn until that condition holds; no argument shows the goal, and clear, stop or cancel drops it. |
| /heapdump | Write a heap snapshot | Hidden from the menu, so type it in full; share only the -diagnostics.json file, since the snapshot holds credentials. |
| /help | Show help and commands | |
| /hooks | List configured hooks | |
| /ide | Manage IDE integrations | |
| /import [codex|gemini|cursor] [--dry-run] [--yes] | Import other agents' config | Carries over instructions, MCP servers, custom commands, subagents and skills; --dry-run previews and --yes skips the picker. |
| /init | Create a CLAUDE.md | CLAUDE_CODE_NEW_INIT=1 adds an interactive flow covering skills, hooks and personal memory; found Codex or Gemini config can be imported. |
| /insights | Usage insights report | Writes an HTML report on your projects, habits, failure points and suggested features; cloud sessions cannot run it. |
| /install-github-app | Install the GitHub App | Supports github.com repositories only; a GitLab or Bitbucket remote prints a notice instead of starting setup. |
| /install-slack-app | Install the Slack app | Setup finishes in the browser through an OAuth flow. |
| /keybindings | Open keybindings file | |
| /list-agents | List messageable agents | Gives the name to address each subagent, teammate or session by; needs cross-session messaging. Alias /peers. |
| /login | Sign in to Anthropic | |
| /logout | Sign out of Anthropic | |
| /loop [interval] [prompt] | Repeat a prompt on a loop | Bundled skill; without an interval Claude paces itself, and without a prompt it runs the maintenance prompt or loop.md. |
| /mcp [reconnect (<server>|all)|enable|disable [<server>|all]] | Manage MCP servers | No argument opens the server list; reconnect all retries servers that failed or need auth, and under -p it prints a status summary. |
| /memory | Edit memory and CLAUDE.md | Also switches auto memory on or off and shows the entries it has saved. |
| /mobile | Mobile app QR code | Scan it to get the Claude mobile app on your phone; aliases /ios and /android. |
| /model [model] | Switch model | Saves the choice as your default; in the picker, press s to switch for this session only, and arrows adjust effort. |
| /output-style [style] | Change output style | Lists the available styles, or switches straight to one such as concise; needs v2.1.269+. |
| /passes | Share free Claude Code weeks | Visible only when your account qualifies for passes. |
| /permissions | Manage tool permission rules | A dialog to view rules by scope, edit them, manage working directories and review auto mode denials; alias /allowed-tools. |
| /plan [description] | Enter plan mode | A description starts planning that task straight away, as in /plan fix the auth bug. |
| /plugin [subcommand] | Manage plugins | No argument opens the plugin menu; subcommands like list, install, enable or disable act directly. |
| /plugin-authoring | Mod-writing reference | Comes from a plugin bundled with Claude Code that /plugin can disable; Claude also loads it when asked for a mod. |
| /powerup | Interactive feature lessons | Short lessons with animated demos that walk you through Claude Code features. |
| /pr-comments [PR] | PR comments (removed) | No longer available since v2.1.91; simply ask Claude to show the comments on a pull request. |
| /privacy-settings | Privacy settings | Available only to Pro and Max plan subscribers. |
| /radio | Lo-fi radio stream | Opens Claude FM in your browser; without a browser, the stream URL is printed instead. |
| /rate-limit-options | Usage-limit options menu | Offers waiting and auto-continuing when the limit resets, buying usage credits, or upgrading; needs a claude.ai subscription. |
| /recap | One-line session summary | Produces the recap on demand, separate from the automatic one shown after you have been away. |
| /release-notes | Browse release notes | Pick one version or all; the notes show in the transcript but stay out of Claude's context. |
| /reload-plugins [--force] | Reload active plugins | Applies pending plugin changes without a restart; --force proceeds even when MCP tool changes would invalidate the prompt cache. |
| /reload-skills | Rescan skill directories | Picks up skills added or edited on disk mid-session, then reports the totals and what changed. |
| /remote-control | Enable Remote Control | Lets you continue this session from claude.ai; requires a claude.ai subscription. Alias /rc. |
| /remote-env | Default cloud environment | |
| /rename [name] | Rename the session | Without a name one is generated from the conversation; names show on the prompt bar and are capped at 200 characters. |
| /resume [session] | Resume a conversation | Takes an ID or name, or opens a picker where bg marks background sessions; a running one gets attached. Alias /continue. |
| /review [low|medium|high|xhigh|max|ultra] [--fix] [--comment] [--max-findings n|all|default] [pr#|branch|path] | Alias of /code-review | Same levels and flags, e.g. /review 1234; with no level it reuses the last one from low to max you typed. |
| /rewind | Rewind to a checkpoint | Rolls back conversation, code or both, or summarizes from a chosen message; aliases /checkpoint and /undo. |
| /run | Run your app | Bundled skill that launches and drives the app so you see the change working, not just tests passing. |
| /run-skill-generator | Generate a run skill | Bundled skill writing a per-project skill so /run and /verify know how to build and launch your app. |
| /sandbox | Toggle sandbox mode | Only offered on platforms that support sandboxing. |
| /schedule [description] | Manage cloud routines | Claude sets up, updates, lists or runs routines with you in conversation; alias /routines. |
| /scroll-speed | Mouse scroll speed | Fullscreen rendering only, and unavailable in JetBrains IDE terminals. |
| /security-review | Security review of branch | Compares your branch with origin's default branch for injection, auth and data-exposure risks; needs an origin remote. |
| /setup-bedrock | Amazon Bedrock setup wizard | Type it in full: the menu hides it unless CLAUDE_CODE_USE_BEDROCK=1; covers auth, region and model pins. |
| /setup-vertex | Agent Platform setup wizard | Shown in the menu only with CLAUDE_CODE_USE_VERTEX=1; sets Google Cloud auth, project, region and model pins. |
| /simplify [target] | Clean up changed code | Bundled skill: four parallel review agents check reuse, simplicity, efficiency and abstraction level, but not correctness bugs. |
| /skill-doctor | Skill cost and usage | Reveals each skill's context cost and how often it runs, to spot skills worth turning off; needs v2.1.252+. |
| /skills | List skills | Type to filter, t sorts by token count, and Space or Enter changes whether Claude and the / menu see a skill. |
| /slides [brief] | Make a slide deck | Bundled skill that fills a Claude Slides artifact from your brief; needs v2.1.265+ and artifacts on your account. |
| /stats | Alias of /usage | Opens the usage screen directly on its Stats tab. |
| /status | Status tab in Settings | Shows version, model, account and connectivity, and works while Claude is responding. |
| /statusline | Configure the status line | Describe the line you want, or run it bare to derive one from your shell prompt. |
| /stickers | Order stickers | |
| /stop | Stop a background session | Keeps the transcript and any worktree; /exit or the left arrow detaches without stopping. |
| /subtask <task> | Spawn a forked subagent | The subagent inherits the whole conversation and reports back here when done, while you carry on working. |
| /tasks | Manage background work | Includes subagents that already finished; also available as /bashes. |
| /team-onboarding | Team onboarding guide | Builds a markdown guide from 30 days of your sessions, commands and MCP usage for a teammate to paste in. |
| /teleport | Pull in a cloud session | Pick a session and its branch and conversation are fetched here; needs a claude.ai subscription. Alias /tp. |
| /terminal-setup | Terminal newline setup | Adds a Shift+Enter newline binding in VS Code, Cursor, Alacritty or Zed, or Option+Enter in Apple Terminal. |
| /theme | Change color theme | Includes auto, light, dark, colorblind-friendly and ANSI themes plus custom ones from ~/.claude/themes/ or plugins. |
| /tui [default|fullscreen] | Switch UI renderer | Relaunches with the conversation intact; fullscreen is the flicker-free renderer, and no argument prints the current one. |
| /ultraplan <prompt> | Cloud planning (removed) | No longer available: use plan mode, which replaced this browser-reviewed cloud planning command. |
| /ultrareview [PR or branch] | Deep cloud code review | Alias of /code-review ultra: a multi-agent review in a cloud sandbox; Pro and Max include 3 free runs. |
| /update-config [request] | Edit settings by request | Bundled skill: describe a change such as an allowed command, env variable or hook and Claude edits settings.json. |
| /upgrade | Upgrade your plan | If the browser cannot open, a sign-in prompt appears instead of the URL. |
| /usage | Cost, limits and activity | Paid plans also itemize what draws on plan limits; /cost and /stats open the same screen. |
| /usage-credits | Configure usage credits | Opens billing in the browser; Team and Enterprise users lacking billing rights send their admin a request instead. |
| /verify | Verify a change works | Bundled skill that builds, runs and observes your app rather than trusting tests; runs only when you invoke it. |
| /vim | Vim mode toggle (removed) | Gone since v2.1.92; switch between Vim and Normal editing in /config under Editor mode. |
| /voice [hold|tap|off] | Toggle voice dictation | The hold, tap or off argument picks the dictation mode; a Claude.ai account is required. |
| /web-setup | Connect GitHub for cloud | Uses the credentials of your local gh CLI to link your GitHub account. |
| /workflow-authoring | Workflow script reference | Bundled skill covering the script API, resume behavior and examples; load it before hand-editing a saved workflow script. |
| /workflows | Workflow progress view | Lets you pause, resume, keep watching or save workflows, whether still running or done. |
Prompt Input Prefixes (5)
| Command / Syntax | What it does | Notes |
|---|---|---|
| / | Run a command or skill | Counts only at the start of a message; typed after a space mid-prompt it just completes a command name. |
| ! | Shell mode | Must start the input; the command skips Claude's approval, and Claude responds once its output lands in the transcript. |
| @ | File path mention | Opens path autocomplete; with cross-session messaging, a letter after it also suggests other live sessions. |
| :name: | Insert an emoji | Two or more letters after a colon open suggestions; the shortcode must start the input or follow a space. |
| ? | Toggle shortcut help | Works on an empty prompt only; with text already typed it inserts a question mark. |
Hook Events (33)
| Command / Syntax | What it does | Notes |
|---|---|---|
| SessionStart | Session begins or resumes | Matcher picks the start type: startup, resume, clear, compact or fork; exit 2 stderr goes to the user only. |
| Setup | One-time setup run | Fires for --init-only, or --init and --maintenance under -p; matcher values init and maintenance, exit codes ignored. |
| UserPromptSubmit | Prompt submitted | No matcher; exit 2 blocks the prompt so Claude never sees it, including on turns Claude Code begins by itself. |
| UserPromptExpansion | Command expands to a prompt | Matcher takes skill or command names; exit 2 stops the expansion before Claude receives it. |
| PreToolUse | Before a tool call | Matcher filters tool names such as Bash, Edit|Write or mcp__.*; exit 2 blocks the call. |
| PermissionRequest | Permission decision needed | Matches tool names; exit 2 is not honored here, so deny through the decision object instead. |
| PermissionDenied | Auto mode denied a call | Matches tool names; the denial already happened, so exit codes are ignored and only the JSON retry flag counts. |
| PostToolUse | After a successful tool call | Matches tool names; exit 2 cannot undo the call but passes stderr to Claude. |
| PostToolUseFailure | After a failed tool call | Matches tool names; on exit 2 Claude sees stderr, since the tool has already failed. |
| PostToolBatch | Parallel tool batch resolved | No matcher; exit 2 halts the agentic loop instead of making another model request. |
| Notification | Notification sent | Matcher filters the notification type, e.g. permission_prompt, idle_prompt or agent_completed; exit codes are ignored. |
| MessageDisplay | Assistant text shown | Display-only with no matcher; on exit 2 the original text is still displayed. |
| SubagentStart | Subagent spawned | Matcher takes agent types such as general-purpose, Explore, Plan or custom names; exit 2 stderr lands in the subagent's own transcript. |
| SubagentStop | Subagent finished | Same matcher values as SubagentStart; exit 2 keeps the subagent working instead of stopping. |
| TaskCreated | Task being created | Fires on TaskCreate with no matcher; exit 2 rolls the new task back. |
| TaskCompleted | Task marked complete | No matcher; exit 2 keeps the task from being marked done. |
| Stop | Claude finished responding | No matcher; exit 2 makes Claude carry on with the conversation instead of stopping. |
| StopFailure | Turn ended by API error | Matcher takes error types like rate_limit, overloaded or server_error; output and exit code are ignored. |
| TeammateIdle | Teammate about to idle | Agent teams only, with no matcher; exit 2 keeps the teammate working. |
| InstructionsLoaded | Instructions file loaded | Fires at session start and on lazy loads; matcher values include session_start, nested_traversal, include and compact. |
| ConfigChange | Config file changed | Matcher takes the source, e.g. user_settings or skills; exit 2 blocks the change except for policy_settings. |
| CwdChanged | Working directory changed | No matcher, so it fires on each change; handy for reloading environment variables with direnv. |
| DirectoryAdded | Directory added mid-session | Matcher values slash_command and register_repo_root; stderr goes only to the debug log. |
| FileChanged | Watched file changed | The matcher lists literal file names to watch, such as .envrc|.env, separated by a pipe. |
| WorktreeCreate | Worktree being created | Takes over from the default git behaviour; any non-zero exit makes the creation fail. |
| WorktreeRemove | Worktree being removed | Covers worktrees a WorktreeCreate hook made; a non-zero exit fails removal if the directory still exists. |
| PreCompact | Before compaction | Matcher manual or auto; exit 2 blocks the compaction. |
| PostCompact | After compaction | Matcher manual or auto; stderr from exit 2 is shown to the user only. |
| PreModelSwitch | Before a model switch | Matcher tests the canonical target model name, e.g. claude-opus-5; exit 2 blocks the switch and shows stderr. |
| PostModelSwitch | After the model changed | Also fires for switches Claude Code makes itself, like restoring the model on resume; it cannot undo the switch. |
| Elicitation | MCP server asks for input | Matcher takes MCP server names; exit 2 declines the request without showing a dialog. |
| ElicitationResult | Elicitation answered | Runs before the reply returns to the server; exit 2 turns the action into a decline. |
| SessionEnd | Session ends | Matcher gives the reason: clear, resume, logout, prompt_input_exit or other; stderr goes to the user only. |
Recipes
Command chains for recurring Claude Code situations, from a filling context window to session cost; the slash commands in them all appear in the tables above.
Keep a long session inside its context window
/context all → /compact [instructions] → /autocompact [auto|<tokens>]
/context draws the window as a grid and points at context-heavy tools and memory bloat; in fullscreen, add all to see the per-item breakdown. /compact takes instructions on what the summary should keep, which beats a generic summary. If automatic compaction arrives at awkward moments, /autocompact sets how full the window may get first (a size such as 500k), and auto returns to the default for your model.
Try another direction without losing the conversation
/branch [name] → /resume [session]
/branch moves you into a copy of the conversation at this point and keeps the original, which /resume brings back. /fork starts a background copy while you keep working here, and /subtask hands a task to a forked subagent whose result returns to this conversation. If an attempt goes wrong, /rewind rolls code and conversation back to a checkpoint.
Check a change before you ship it
/diff → /code-review high --fix → /security-review
/diff shows Claude's edits next to anything else you have not committed. /code-review looks for correctness bugs at the level you pick, and --fix applies what it finds; pass a PR number to review a pull request instead. /security-review then checks the branch against origin's default branch for injection, auth and data-exposure risks.
Feed a command's real output back to the model
! npm test
The ! prefix runs a shell command without going through Claude and adds the output to the session. Claude answers it on its own, so one line gets you an explanation of the failing tests, at the cost of a normal prompt; set respondToBashCommands to false to add output without a reply.
See what a session is costing
/usage → /rate-limit-options
/usage shows session cost, plan limits and activity; /cost and /stats are aliases, and /stats opens on the Stats tab. When a claude.ai usage limit blocks a request, /rate-limit-options lists the ways to continue: wait for the reset, add usage credits or upgrade.
Point Claude at an outcome instead of a single instruction
/plan [description] → /goal [condition|clear] → /loop [interval] [prompt]
/plan enters plan mode so you can approve the approach first. /goal keeps Claude working across turns until a condition is met, and /goal clear drops it early. /loop repeats a prompt while the session stays open, for example every 5m; leave out the interval and Claude paces itself.
Frequently Asked Questions
How do I see which slash commands my Claude Code version supports?
Type / in the Claude Code prompt, or / plus a few letters to filter. Commands that are not available to you are left out of the menu. A few commands stay out of the menu until you type the full name, for example /heapdump, and /setup-bedrock or /setup-vertex before their provider variable is set.
What do the square and angle brackets in the command column mean?
They follow the vendor's notation: <path> marks a required argument and [name] an optional one; type the value, not the brackets. A bar separates choices, so /fast [on|off] takes on, off or nothing.
What is the difference between / and ! at the start of a message?
/ runs a command or skill. ! enters shell mode: the command runs directly, its output is added to the session, and Claude can respond to it.
Which hook events can stop an action with exit code 2?
PreToolUse, UserPromptSubmit, UserPromptExpansion, PostToolBatch, Stop, SubagentStop, TeammateIdle, TaskCreated, TaskCompleted, ConfigChange, PreCompact, PreModelSwitch, Elicitation and ElicitationResult block or reverse what they fire on; WorktreeCreate and WorktreeRemove fail on any non-zero exit. For PermissionRequest, deny through the JSON decision object instead.
Where did the Claude Code keyboard shortcuts go?
Keyboard shortcuts and CLI startup flags are on the Claude Code shortcuts page. This page covers in-session commands, prompt input prefixes and hook events.
Sources and checks
- Claude Code Docs — Commands (current for Claude Code v2.1.296) Built-in slash commands: the "All commands" table, signatures with their [optional] and <required> arguments
- Claude Code Docs — Hooks reference (current for Claude Code v2.1.296) Hook events: the lifecycle table, the matcher table and the exit code 2 table
- Claude Code Docs — Interactive mode (current for Claude Code v2.1.296) Prompt input prefixes: the "Quick commands" table, shell mode and emoji shortcodes
- Claude Code Docs — Changelog (newest entry v2.1.296, October 9, 2026) Version check for the docs above; the merge of /cost and /stats into /usage
Every row on this page was matched to a line in the documentation above on . A command or parameter that could not be found there is not listed.
AI tools change their commands, parameters and model names between releases: the vendor's current documentation and the version you run have the last word. Found a wrong or missing entry? Tell us and it gets checked against the source. How we verify describes the process and its limits.
Related prompt references
Cursor (72) · OpenAI Codex CLI (49) · Aider (46) · All AI prompt references