Add /autopilot slash command for quick tasks in any session

New slash command at ~/.claude/commands/autopilot.md lets users run
autopilot tasks without starting a dedicated agent session:

  /autopilot deploy this to Vercel
  /autopilot set up Supabase with auth

The command injects autopilot instructions for that single task —
plan, execute, log, report — then the user continues coding normally.

Agent mode (claude --agent autopilot) remains for full multi-service
orchestrations. Both share the same guardian, keychain, and logging.

Updated installer to install the slash command automatically.
Updated uninstaller to remove it on cleanup.

Author: rish-e

README.md

@@ -173,33 +173,37 @@ cd autopilot
 
 ## Usage
 
-```bash
-# Start Autopilot (recommended — Guardian provides the safety rails)
-claude --agent autopilot --dangerously-skip-permissions
-
-# Deploy and host
-> Deploy this to Vercel with environment variables from Supabase
+### Two ways to use Autopilot
 
-# Databases
-> Set up Supabase with user auth tables, API keys, and usage tracking
+**Slash command** — use from any Claude Code session (recommended for most tasks):
 
-# Payments
-> Configure Stripe payments with webhooks for subscription billing
+```bash
+# Inside any Claude Code session, type:
+/autopilot deploy this to Vercel with environment variables from Supabase
+/autopilot set up Supabase with user auth tables and API keys
+/autopilot configure Stripe payments with webhooks
+/autopilot create a Cloudflare R2 bucket for image storage
+```
 
-# Infrastructure
-> Create a Cloudflare R2 bucket for image storage and set up a Worker for resizing
+**Agent mode** — dedicated session for big multi-service orchestrations:
 
-# CI/CD
-> Set up GitHub Actions to run tests on PR and deploy to Vercel on merge to main
+```bash
+# Start a full Autopilot session
+claude --agent autopilot --dangerously-skip-permissions
 
-# Multiple services at once
 > I need this running in production with a Postgres database, Stripe payments, and Sentry monitoring
-
-# Services it's never seen before
-> Set up Resend for transactional emails and Upstash for Redis caching
 ```
 
-Autopilot figures out the rest. If it's a service it hasn't seen before, it researches the docs, creates a registry file, installs the CLI, and keeps going. First time with a service, it asks for your login credentials once. Every subsequent interaction is fully autonomous (except 2FA codes — those need your phone).
+### When to use which
+
+| Situation | Use |
+|-----------|-----|
+| Quick deploy, get an API key, install a service | `/autopilot` slash command |
+| Full project setup from scratch | Agent mode |
+| Mid-coding infrastructure task | `/autopilot` slash command |
+| Multi-service orchestration (5+ services) | Agent mode |
+
+Autopilot figures out the rest. If it's a service it hasn't seen before, it researches the docs, creates a registry file, installs the CLI, and keeps going. Your primary credentials handle signups automatically. Every subsequent interaction is fully autonomous (except 2FA codes — those need your phone).
 
 ### Execution Log
 
@@ -250,8 +254,10 @@ If something breaks midway, open the log to see exactly what happened and where.
     github.md             # GitHub: repos, PRs, issues, Actions
     cloudflare.md         # Cloudflare: Workers, R2, KV, DNS
     razorpay.md           # Razorpay: payments, subscriptions, webhooks
+  commands/
+    autopilot.md          # /autopilot slash command (installed to ~/.claude/commands/)
   agent/
-    autopilot.md          # The agent definition (installed to ~/.claude/agents/)
+    autopilot.md          # Full agent definition (installed to ~/.claude/agents/)
 
 # Per-project (created automatically when Autopilot runs):
 your-project/.autopilot/

commands/autopilot.md

@@ -0,0 +1,42 @@
+You are now operating in **Autopilot mode** for this task only. Execute the user's request autonomously using the full Autopilot system.
+
+## How to operate
+
+1. **Analyze** the task — what services, CLIs, credentials, and steps are needed
+2. **If simple** (single service, 1-3 steps): execute immediately with `[1/N]` status updates
+3. **If complex** (multi-service, 4+ steps): present a numbered plan, wait for "proceed", then execute all steps without pausing
+4. **Log everything** to `{project}/.autopilot/log.md` (create if needed, add `.autopilot/` to `.gitignore`)
+5. **Report** the result when done
+
+## Rules
+
+- **Never pause between steps** to ask "what next?" — execute to completion or until genuinely blocked
+- **Credentials**: check `~/MCPs/autopilot/bin/keychain.sh has {service} {key}` first. If missing, use primary credentials (`keychain.sh get primary email/password`) to sign up or log in via Playwright browser automation
+- **Never expose credentials** in output, logs, or files. Use subshell expansion: `"$(keychain.sh get ...)"`
+- **CLI over browser**: use CLI tools with stored tokens whenever possible
+- **Service registries**: read from `~/MCPs/autopilot/services/{service}.md` for how to interact with each service. If none exists, research the docs (WebSearch), create one, and continue
+- **Safety**: the Guardian hook blocks dangerous commands automatically. Don't try to work around it
+- **Log account activity**: mark ACCOUNT CREATED, LOGGED IN, TOKEN STORED in the execution log
+
+## Execution Log Format
+
+Append to `{project}/.autopilot/log.md`:
+
+```markdown
+## Session: {YYYY-MM-DD HH:MM} — {task description}
+
+| # | Time | Action | Level | Service | Result |
+|---|------|--------|-------|---------|--------|
+```
+
+## Decision Levels
+
+| Level | Action |
+|-------|--------|
+| L1 | Just do it, brief note |
+| L2 | Do it, notify |
+| L3 | Ask first |
+| L4 | Must ask (money, messages, publishing) |
+| L5 | Escalate (2FA, CAPTCHA) |
+
+Now execute the following task: $ARGUMENTS

install.sh

@@ -213,8 +213,10 @@ else
 fi
 
 # Create directories
-mkdir -p "$INSTALL_DIR"/{bin,config,services}
+COMMANDS_DIR="$HOME/.claude/commands"
+mkdir -p "$INSTALL_DIR"/{bin,config,services,commands}
 mkdir -p "$AGENT_DIR"
+mkdir -p "$COMMANDS_DIR"
 
 # Copy core files
 cp -f "$SOURCE_DIR/bin/keychain.sh" "$INSTALL_DIR/bin/"
@@ -247,6 +249,10 @@ ok "Service registry installed"
 cp -f "$SOURCE_DIR/agent/autopilot.md" "$AGENT_DIR/autopilot.md"
 ok "Agent definition installed at $AGENT_DIR/autopilot.md"
 
+# Copy slash command
+cp -f "$SOURCE_DIR/commands/autopilot.md" "$COMMANDS_DIR/autopilot.md"
+ok "Slash command installed (/autopilot)"
+
 # Make scripts executable
 chmod +x "$INSTALL_DIR/bin/"*.sh
 ok "Scripts made executable"
@@ -391,7 +397,8 @@ echo ""
 echo -e "${GREEN}${BOLD}Autopilot installed successfully!${NC}"
 echo ""
 echo "  Platform: $PLATFORM"
-echo "  Start it: claude --agent autopilot --dangerously-skip-permissions"
+echo "  Quick use: /autopilot <task>  (from any Claude Code session)"
+echo "  Full mode: claude --agent autopilot --dangerously-skip-permissions"
 echo ""
 echo "  What's installed:"
 echo "    Agent:     $AGENT_DIR/autopilot.md"

uninstall.sh

@@ -10,6 +10,7 @@ NC='\033[0m'
 
 INSTALL_DIR="$HOME/MCPs/autopilot"
 AGENT_FILE="$HOME/.claude/agents/autopilot.md"
+COMMAND_FILE="$HOME/.claude/commands/autopilot.md"
 SETTINGS_FILE="$HOME/.claude/settings.json"
 SETTINGS_LOCAL="$HOME/.claude/settings.local.json"
 
@@ -23,6 +24,12 @@ if [ -f "$AGENT_FILE" ]; then
     echo -e "${GREEN}[OK]${NC} Removed agent definition"
 fi
 
+# Remove slash command
+if [ -f "$COMMAND_FILE" ]; then
+    rm "$COMMAND_FILE"
+    echo -e "${GREEN}[OK]${NC} Removed /autopilot slash command"
+fi
+
 # Remove guardian hook from settings.json
 if [ -f "$SETTINGS_FILE" ]; then
     if jq -e '.hooks.PreToolUse' "$SETTINGS_FILE" &>/dev/null; then