diff --git a/lib/prism/cli.rb b/lib/prism/cli.rb
index 42f2f0e..c47a564 100644
--- a/lib/prism/cli.rb
+++ b/lib/prism/cli.rb
@@ -23,10 +23,191 @@ module Prism
       case command
       when 'new' then new_app(rest.first || raise(UsageError, 'usage: prism new APP_NAME'))
       when 'generate', 'g' then generate(rest)
+      when 'jobs' then jobs(rest)
+      when 'assets:precompile' then assets_precompile(rest)
+      when 'console', 'c' then console(rest)
+      when 'routes' then routes(rest)
+      when 'db:migrate' then db_migrate(rest)
+      when 'db:rollback' then db_rollback(rest)
+      when 'db:status' then db_status(rest)
+      when 'db:seed' then db_seed(rest)
       else raise UsageError, "unknown command #{command.inspect}"
       end
     end
 
+    # Applies every pending migration directly against a PStore file, the
+    # same work `Orm.migrator(...).migrate!` does when a generated app.rb
+    # boots - but as a standalone step that does not require loading the
+    # rest of the application (no controllers, no server), mirroring `rails
+    # db:migrate`. Useful on its own for apps that do not auto-migrate at
+    # boot, and it is the only way to inspect/rewind migrations (db:status,
+    # db:rollback below) without booting an app that would immediately
+    # re-apply everything pending anyway.
+    def db_migrate(args, out: $stdout)
+      migrator = build_migrator(parse_db_options(args))
+      applied = migrator.migrate!
+      if applied.empty?
+        out.puts '[prism db:migrate] no pending migrations'
+      else
+        applied.each { |version| out.puts "[prism db:migrate] migrated #{version}" }
+      end
+      applied
+    end
+
+    # Reverts the most recently applied migration(s), like `rails
+    # db:rollback STEP=n` - `--step=N` picks how many to undo (default 1).
+    def db_rollback(args, out: $stdout)
+      options = parse_db_options(args)
+      migrator = build_migrator(options)
+      reverted = migrator.rollback!(options[:step])
+      if reverted.empty?
+        out.puts '[prism db:rollback] nothing to roll back'
+      else
+        reverted.each { |version| out.puts "[prism db:rollback] rolled back #{version}" }
+      end
+      reverted
+    end
+
+    # Lists every migration file found under --migrations next to whether
+    # it has been applied to --database, like `rails db:migrate:status`.
+    def db_status(args, out: $stdout)
+      migrator = build_migrator(parse_db_options(args))
+      status = migrator.status
+      rows = status.map { |entry, applied| [applied ? 'up' : 'down', entry.version, entry.name] }
+      header = %w[STATUS VERSION NAME]
+      widths = header.each_index.map { |i| ([header[i].length] + rows.map { |row| row[i].length }).max }
+      out.puts format_row(header, widths)
+      rows.each { |row| out.puts format_row(row, widths) }
+      status
+    end
+
+    # Loads the application (via --require, same flag shape as `jobs work`/
+    # `console`) and then `load`s --file (default db/seeds.rb) as plain Ruby,
+    # like `rails db:seed`. Deliberately `load` rather than `require`: a
+    # seeds file has no meaningful "already loaded" state of its own (unlike
+    # a model or controller file, it is not a class/module definition but a
+    # sequence of side effects), so a second `db:seed` run - or a console/
+    # test process that seeds more than once - must re-execute it, not
+    # silently no-op the way `require` would after the first call already
+    # loaded that exact path.
+    #
+    # Idempotency of what the seeds file actually does (find_or_create_by
+    # vs. create!, etc.) is the app's concern, not this command's - same
+    # division of responsibility Rails draws. A missing seeds file is not an
+    # error: plenty of apps never have one, and `prism new` does not
+    # generate a stub, so failing here would make db:seed unusable as a
+    # step in a deploy script unless every app remembered to create an empty
+    # placeholder first.
+    def db_seed(args, out: $stdout)
+      options = parse_seed_options(args)
+      options[:require].each { |path| require File.expand_path(path, @root) }
+
+      seeds_path = File.expand_path(options[:file], @root)
+      unless File.exist?(seeds_path)
+        out.puts "[prism db:seed] no seeds file at #{options[:file]}"
+        return false
+      end
+
+      load seeds_path
+      out.puts "[prism db:seed] seeded from #{options[:file]}"
+      true
+    end
+
+    # Lists every route registered on a loaded Application subclass, like
+    # `rails routes`: NAME, VERB, PATH and the controller#action (or
+    # "(block)" for routes defined inline rather than via `to:`) it
+    # dispatches to, in registration order (the same order the router tries
+    # them in, which matters when patterns overlap). Mounted Rack apps
+    # (`mount`) are listed separately underneath since they do not carry a
+    # single verb/action - only a prefix and whatever app answers under it.
+    def routes(args, out: $stdout)
+      options = parse_routes_options(args)
+      loaded_before = Application.subclasses
+      options[:require].each { |path| require File.expand_path(path, @root) }
+      newly_loaded = Application.subclasses - loaded_before
+
+      app_class = resolve_app_class(options[:app], newly_loaded)
+      print_routes(app_class.router, out)
+      app_class
+    end
+
+    # A minimal `rails console`-style REPL: loads the application (via
+    # --require, same flag shape as `jobs work`) and evaluates one line of
+    # Ruby at a time against a binding that survives across lines, so a
+    # local variable assigned on one line is still visible on the next -
+    # e.g. `post = Post.first` then `post.title`. Deliberately single-line
+    # (no heredoc/multi-line statement support) and has no history/tab
+    # completion, unlike a real irb - this is a REPL, not a terminal editor;
+    # it reads whatever `input` yields line by line, which makes it just as
+    # easy to drive from a real TTY as from a StringIO in tests.
+    def console(args, input: $stdin, out: $stdout)
+      options = parse_console_options(args)
+      options[:require].each { |path| require File.expand_path(path, @root) }
+
+      session = Object.new
+      session_binding = session.instance_eval { binding }
+
+      prompt = options[:prompt]
+      out.print(prompt)
+      input.each_line do |line|
+        line = line.chomp
+        stripped = line.strip
+        if %w[exit quit].include?(stripped)
+          break
+        elsif stripped.empty?
+          out.print(prompt)
+          next
+        end
+
+        begin
+          result = session_binding.eval(line, '(console)')
+          out.puts "=> #{result.inspect}"
+        rescue StandardError, ScriptError => e
+          out.puts "Error: #{e.class}: #{e.message}"
+        end
+        out.print(prompt)
+      end
+      out.puts
+      nil
+    end
+
+    # Runs Assets::Compiler directly - fingerprints every file under
+    # --source into --output plus a manifest.json - without booting the
+    # rest of the application (no Orm.connect, no migrations, no server).
+    # This is the missing deploy-time step Application#enable_assets never
+    # covered on its own: enable_assets compiles as a side effect of
+    # loading app.rb, which also means the ONLY way to precompile assets
+    # before this was to boot the entire app (database connection and all)
+    # just to get the compiled files onto disk. Mirrors `rake
+    # assets:precompile` in Rails: a standalone step meant to run once
+    # during deploy, before the server process starts.
+    def assets_precompile(args, out: $stdout)
+      options = parse_assets_precompile_options(args)
+      source = File.expand_path(options[:source], @root)
+      output = File.expand_path(options[:output], @root)
+      manifest = Assets::Compiler.new(source: source, output: output).compile!
+      out.puts "[prism assets:precompile] compiled #{manifest.to_h.size} file(s) into #{output}"
+      manifest
+    end
+
+    # Runs a Jobs::Worker as a long-lived daemon: polls the queue, drains
+    # every ready job, sleeps `interval` seconds, repeats until SIGINT/SIGTERM
+    # (or forever with --once/--max-cycles for scripted/finite runs). Returns
+    # the number of poll cycles executed, mainly so callers/tests can assert
+    # on it without needing a real infinite loop.
+    def jobs_work(args, out: $stdout)
+      options = parse_jobs_work_options(args)
+      options[:require].each { |path| require File.expand_path(path, @root) }
+
+      pidfile = options[:pidfile] && File.expand_path(options[:pidfile], @root)
+      write_pidfile(pidfile) if pidfile
+
+      queue = Jobs::Queue.new(File.expand_path(options[:file], @root))
+      worker = Jobs::Worker.new(
+        queue,
+        queue_name: options[:queue_name],
+    end
+
     def new_app(name)
       app_dir = File.join(@root, name)
       raise UsageError, "#{app_dir} already exists" if Dir.exist?(app_dir)