What each one does, a real scenario from our world (Ty, Alex, the VPS, Lucidya,
the dashboard), and the one best-practice tip. The live list is always /help in a
session — this is the map with examples.
Example (us) = the highlighted column: a concrete scenario from our actual setup showing when & how you'd reach for that command.
| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
claude | Start interactive session in current folder | cd ~/openclaw-vault then claude — the session already knows the vault's CLAUDE.md and Lucidya files | cd into the right folder first |
claude "prompt" | Start with an opening instruction | claude "audit dashboard_v3.py for taste violations" jumps straight in | Quote the whole prompt |
claude -c | Continue most recent conversation here | You closed the terminal mid-Lucidya-deck last night — claude -c drops you right back | Fastest way back into context |
claude -r | Picker of past sessions to resume | Want the dashboard session from 3 days ago, not last night's deck | claude -r <id> jumps to one |
claude -p "prompt" | Run once, print, exit (no UI) | ssh VPS 'tail -200 gateway.log' \| claude -p "what's crashing?" to triage Alex from a script | Pipe logs straight in |
claude --model <name> | Launch on a specific model | --model claude-opus-4-8 for a hard Lucidya strategy pass | Opus reasoning · Haiku trivial |
claude --add-dir <path> | Add an extra working folder | In a client repo, --add-dir ~/openclaw-vault so it sees code and the SOPs | Add vault + project together |
claude --permission-mode plan | Start in Plan mode | Big refactor of the 500KB dashboard_v3.py — plan before touching it | Or just Shift+Tab in-session |
claude --effort <level> | Reasoning budget (low→max) | --effort max for the gateway dual-unit bug; low for a copy fix | Match effort to difficulty (cost) |
claude --dangerously-skip-permissions | Skip all permission prompts | A trusted batch that regenerates 6 deck variants in a scratch dir, unattended | Never near prod, VPS, or deletes |
claude doctor | Diagnose install health | Claude Code acting weird on the MacBook — run this before guessing | First thing when broken |
claude mcp | Configure MCP tool servers | Re-wire ClickUp or Supabase after the connection drops | claude mcp list to see them |
claude --help | Full flag list for your version | Forget the resume flag — claude -h is authoritative | The real list for YOUR build |
--allowedTools/--disallowedTools
(pre-approve/block), --append-system-prompt, --agents '{...}' (inline
subagents), --session-id, --fork-session, --output-format json,
--fallback-model, --from-pr, --debug, --bare./ for the menu.| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/clear | Wipe context, fresh start | Finish the Lucidya deck, switch to debugging Alex's cron — /clear first so deck context doesn't leak in | Use constantly — free quality boost |
/compact [hint] | Summarize to free context | 3-hour dashboard session slowing — /compact keep the card-labeling fixes | Hint what to keep |
/context | Show what's loaded in the window | Answers gone vague — check if old files are eating the window | Check before a big compact |
/resume | Reload a previous conversation | Switch back to the EOM-report session | — |
/export | Export the conversation | Save tonight's dashboard-audit session into the vault as a handoff for Alex | Good for handoffs |
| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/init | Scan project, generate CLAUDE.md | New client repo — /init drafts the CLAUDE.md, then you hand-edit | Then edit it — #1 quality lever |
/memory | Edit your CLAUDE.md memory | Add "Lucidya budget is $20K not $18K" so it never drifts again | Project file · or ~/.claude global |
/model | Switch active model | Drop to Sonnet to crank boilerplate, back to Opus for the hard part | Opus hard · Haiku trivial |
/permissions | Allow/deny rules for tools | You keep approving ssh alexops@… — allowlist it | Allowlist safe repeats |
/hooks | Auto-run scripts on events | "After every .py edit, run the dashboard syntax check" — that's a hook | Automation lives here, not memory |
/agents | Create/manage subagents | An "Explore" subagent sweeps the vault for every Lucidya doc without clogging main context | Keeps big sweeps isolated |
/mcp | Manage tool servers & auth | ClickUp returns empty — /mcp to re-auth | Re-auth when a tool returns empty |
/config | Open settings UI | Switch theme or default model | — |
/terminal-setup | Fix terminal keys (Shift+Enter) | First time on a new machine so multiline prompts work | Run once per machine |
| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/status | Version, account, model, connections | "What model am I on, is MCP connected?" | First stop when off |
/cost | Token + cost this session | Long Lucidya session — check spend given the ~$50/day API pressure | Watch on long sessions |
/doctor | Health-check the install | Tools failing for no clear reason | Same as claude doctor |
/help | List all commands right now | Forget what's installed — this is the truth for your build | The live source of truth |
| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/review | Review a pull request | Review a dashboard PR before it merges | Point at the PR |
/security-review | Security review of branch changes | Before deploying anything touching the VPS or client data | Run before shipping auth/data |
/help shows the current set.| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/code-review | Reviews diff for bugs + cleanups | After editing dashboard_v3.py, /code-review high before you restart the service | ultra=cloud(billed) · --fix · --comment |
/simplify | Cleans changed code (no bug hunt) | The card-label fix works but is messy — tidy it | Pair with /code-review for bugs |
/verify | Runs the app to confirm a change works | Fixed a dashboard card — /verify actually loads it and checks | Beats "looks right" |
/run | Launches the app to see a change live | See the Lucidya deck render in the browser | Screenshot the result |
/deep-research | Fan-out web research, cited | "Current consensus on AI design 'taste' prompts?" — what we did tonight | Give a specific question |
/loop [interval] | Repeat a task on a schedule | /loop 15m to poll whether Alex started the dashboard audit overnight | Omit interval to self-pace |
/schedule | Scheduled remote agents (cron) | "Every morning 8am, summarize Alex's overnight cron failures" | One-time too ("at 3pm…") |
| Command | What it does | Example (us) | Best practice |
|---|---|---|---|
/claude-api | Build/debug Claude API + SDK apps | A script that calls Claude to score Lucidya ad copy | When your code calls Claude |
/update-config | Configure harness (hooks, perms, env) | "From now on, after I edit any .py, auto-run a syntax check" | "Whenever X, do Y" |
/fewer-permission-prompts | Auto-build an allowlist from history | Drowning in SSH/curl approvals — scans and allowlists safe repeats | — |
| Command | What it does | Example (us) | When |
|---|---|---|---|
/alex-bootstrap | Full Alex context — takeover | Gateway's down and you need to operate her infra yourself | Alex offline |
/alex-ops | VPS infrastructure management | Restart the gateway, check preflight drift | Server work |
/alex-cron | Cron job management | Fix the twitter-daily-tweets job that keeps timing out | Scheduled jobs |
/alex-clients | Client context & management | Pull up everything on Lucidya before a call | Client work |
/alex-comms | Discord communication protocol | Post a correctly-routed update to a client channel | Posting for Alex |
/alex-dashboard | Reactiiv Command Center work | The exact dashboard audit running right now | Dashboard work |
| Mode | Behavior | Example (us) |
|---|---|---|
| Normal | Asks permission before edits/commands | Default — safe when touching the live VPS |
| Auto-accept | Applies edits without asking each time | Bulk-fixing 20 card labels in a dashboard you already trust |
| Plan mode | Proposes a plan, changes nothing till you approve | Before any big dashboard_v3.py refactor — the best habit |
| Type | Does | Example (us) |
|---|---|---|
/ | Slash command menu | /clear |
@ | Pull a file into context | @knowledge/playbooks/design/TASTE.md to load the taste rules |
! | Run a shell command, output into chat | ! ssh alexops@100.119.93.118 'uptime' · or ! gcloud auth login |
# | Quick-save a line to memory (verify w/ /memory) | # Lucidya budget is $20K |
| Keys | Does | Example (us) |
|---|---|---|
| Esc | Interrupt mid-response | It's heading down the wrong path on the deck — stop it |
| Esc Esc | Rewind / edit a previous message | Reword your dashboard ask and re-run from there |
| Shift+Tab | Cycle Normal → Auto-accept → Plan | — |
| Ctrl+R | Reverse-search history | Find that long SSH command from earlier |
| Ctrl+L | Clear screen (keeps context) | Tidy a cluttered terminal mid-task |
| Drag / paste image | Add a screenshot to context | Paste the ugly dashboard card and say "fix this" |
CLAUDE.md is your highest-leverage tool. Our whole Alex/Lucidya context lives there
and loads every session. /init to create, /memory to maintain.
Everything else is downstream of this.
/clear between unrelated tasks. Lucidya deck → gateway debugging? Clear first. Stale context = dumb answers. Free.
Plan mode for anything non-trivial. Especially anything touching the live VPS or the 500KB dashboard_v3.py.
Be specific, give references. "Cards like Stripe's — one accent, labeled metrics" beats "make it nicer."
@-mention exact files (@TASTE.md) instead of describing them.
/compact before the wall, with a hint about what to keep.
Allowlist safe repeats (/permissions) — stop approving the same SSH command 40 times.
Hooks for automation, memory for knowledge. "Do X every time Y" needs a hook — memory only informs.
/verify or /run to confirm. Make it prove the dashboard fix actually renders.
/doctor first when something feels broken — and /help is always the truth for what's installed.