diff --git a/spec/ysh-stdlib-testing.test.sh b/spec/ysh-stdlib-testing.test.sh
index 4e5b391adc..c8a0c2f093 100644
--- a/spec/ysh-stdlib-testing.test.sh
+++ b/spec/ysh-stdlib-testing.test.sh
@@ -1,91 +1,88 @@
 ## our_shell: ysh
-## oils_failures_allowed: 5
 
-#### value.Expr test - positional test
+#### Test framework
 
 source --builtin testing.ysh
+setglobal _test_use_color = false
 
-echo 'parens'
-test-expr (42 + 1)
-echo
-
-echo 'brackets'
-test-expr [42 + 1]
-echo
-
-echo 'expr in parens'
-test-expr (^[42 + 1])
-echo
-
-## STDOUT:
-## END
-
-#### value.Expr test - named test
-
-source --builtin testing.ysh
-
-echo 'parens'
-test-named (n=42 + 1)
-echo
-
-echo 'brackets'
-test-named [n=42 + 1]
-echo
-
-echo 'expr in parens'
-test-named (n=^[42 + 1])
-echo
-
-echo 'no value'
-test-named
-echo
-
-## STDOUT:
-## END
-
-#### assert builtin
-
-source --builtin testing.ysh  # get rid of this line later?
-
-var x = 42
-
-# how do you get the code string here?
-
-assert [42 === x]
+test-suite "three bad tests" {
+    test-case "the assertion is false" {
+        assert [1 > 0]
+        assert [1 > 1]
+        assert [1 > 2]
+    }
 
-assert [42 < x]
+    test-case "there is an exception while evaluating the assertion" {
+        assert ["Superman" > "Batman"]
+    }
 
-#assert [42 < x; fail_message='message']
+    test-case "there is an exception outside of any assertion" {
+        error "oops!"
+    }
+}
 
-#assert (^[(42 < x)], fail_message='passed message')
+test-case "one good test case" {
+    assert [1 === 1]
+    assert [2 === 2]
+}
 
-# BUG
-assert [42 < x, fail_message='passed message']
+run-tests
 
 ## STDOUT:
+begin three bad tests
+    test the assertion is false ...
+        assertion FAILED
+    test there is an exception while evaluating the assertion ...
+        assertion ERRORED: 10
+    test there is an exception outside of any assertion ...
+        ERROR: 10
+end
+test one good test case ... ok
+3 / 4 tests failed
 ## END
 
-#### ysh --tool test file
+#### Test framework: nested test suites
 
-cat >mytest.ysh <<EOF
-echo hi
-EOF
-
-# which ysh
+source --builtin testing.ysh
+setglobal _test_use_color = false
 
-# the test framework sets $SH to bin/ysh
-# but ysh is already installed on this machine
+test-case "A" { : }
+test-suite "outer test suite" {
+    test-case "B" { : }
+    test-suite "first inner test suite" {
+        test-case "C" { : }
+    }
+    test-case "D" { : }
+    test-suite "second inner test suite" {
+        test-case "E" { : }
+    }
+    test-case "F" { : }
+}
+test-case "G" { : }
 
-$SH --tool test mytest.ysh
+run-tests
 
 ## STDOUT:
+test A ... ok
+begin outer test suite
+    test B ... ok
+    begin first inner test suite
+        test C ... ok
+    end
+    test D ... ok
+    begin second inner test suite
+        test E ... ok
+    end
+    test F ... ok
+end
+test G ... ok
+7 tests succeeded
 ## END
 
-# Hm can we do this entirely in user code, not as a builtin?
-
-#### Describe Prototype
+#### Test framework: stdout and stderr
 
 source --builtin testing.ysh
+setglobal _test_use_color = false
 
 proc p {
   echo STDOUT
@@ -93,33 +90,41 @@ proc p {
   return 42
 }
 
-describe p {
+test-case "that it prints to stdout and stderr" {
   # each case changes to a clean directory?
   #
   # and each one is numbered?
 
-  it 'prints to stdout and stderr' {
-    try {
-      p > out 2>& err
-    }
-    assert (_status === 42)
-
-    cat out
-    cat err
-
-    # Oh man the here docs are still useful here because of 'diff' interface
-    # Multiline strings don't quite do it
-
-    diff out - <<< '''
-    STDOUT
-    '''
-
-    diff err - <<< '''
-    STDERR
-    '''
+  try {
+    p > out 2> err
   }
+
+  assert [_status === 42]
+  assert [$(<out) === "STDOUT"]
+  assert [$(<err) === "STDERR"]
 }
 
+run-tests
+
 ## STDOUT:
-TODO
+test that it prints to stdout and stderr ... ok
+1 tests succeeded
 ## END
+
+# #### ysh --tool test file
+# 
+# cat >mytest.ysh <<EOF
+# echo hi
+# EOF
+# 
+# # which ysh
+# 
+# # the test framework sets $SH to bin/ysh
+# # but ysh is already installed on this machine
+# 
+# $SH --tool test mytest.ysh
+# 
+# ## STDOUT:
+# ## END
+# 
+# # Hm can we do this entirely in user code, not as a builtin?
diff --git a/stdlib/testing.ysh b/stdlib/testing.ysh
index 7abce14367..b154cfe85a 100644
--- a/stdlib/testing.ysh
+++ b/stdlib/testing.ysh
@@ -1,115 +1,202 @@
-# testing.ysh
-#
-# Usage:
-#   source --builtin testing.sh
-#
-# func f(x) { return (x + 1) }
-#
-# describe foo {
-#   assert (43 === f(42))
-# }
-#
-# if is-main {
-#   run-tests @ARGV   # --filter
-# }
-
-module stdlib/testing || return 0
-
-source --builtin args.ysh
-
-proc assert ( ; cond ; fail_message='default fail message') {
-  echo 'hi from assert'
-
-  = cond
-
-  # I think this might be ready now?
-
-  var val = evalExpr(cond) 
-
-  echo
-  echo 'value'
-  = val
-  pp line (val)
-
-  = fail_message
-
-  if (val) {
-    echo 'OK'
-  } else {
-    var m = evalExpr(fail_message) 
-    echo "FAIL - this is where we extract the string - $m"
-  }
+# TODO:
+# - [x] Renaming a bit for "privacy"
+# - [x] Renaming for public interface: test-case, test-suite, assert (subject to change)
+# - [x] Rename to use "-"s for procs.
+# - [x] Make test-case and test-suite take a word arg so they don't need parens
+# - [x] error if you try to re-enter run-tests
+# - [x] testing testing: assert fail, exn in assert, exn outside of assert
+# - [-] check for both kinds of exceptions (_status and _error)
+# - [x] tests for tests
+# - [ ] question: naming for variables?
+# - [ ] --tool test support (& spec test case for it)
+# - [ ] Turn the assert expression into a string, to print if it fails
+# - [ ] Use ctx instead of globals, wherever possible
+# - [ ] Merge _status and _error. Every exception should set _error, with the
+#       status stored on _error.status. Use _error instead of the icky global
+#       _assertion_result
+
+##########
+# Colors #
+##########
+
+const _BOLD = $'\e[1m'
+const _RED = $'\e[31m'
+const _GREEN = $'\e[32m'
+const _YELLOW = $'\e[33m'
+const _PURPLE = $'\e[35m'
+const _CYAN = $'\e[36m'
+const _RESET = $'\e[0;0m'
+
+func _red(text) {
+    if (_test_use_color) {
+        return ("${_BOLD}${_RED}${text}${_RESET}")
+    } else {
+        return (text)
+    }
 }
-
-proc test-assert {
-  var x = 42
-  assert [42 === x]
+func _yellow(text) {
+    if (_test_use_color) {
+        return ("${_YELLOW}${text}${_RESET}")
+    } else {
+        return (text)
+    }
 }
-
-proc test-expr ( ; expr ) {
-  echo 'expr'
-  pp line (expr)
+func _green(text) {
+    if (_test_use_color) {
+        return ("${_YELLOW}${text}${_RESET}")
+    } else {
+        return (text)
+    }
 }
-
-proc test-named ( ; ; n=^[99] ) {
-  echo 'n'
-  pp line (n)
+func _cyan(text) {
+    if (_test_use_color) {
+        return ("${_YELLOW}${text}${_RESET}")
+    } else {
+        return (text)
+    }
 }
 
-# What happens when there are duplicate test IDs?
-#
-# Also I think filter by "$test_id/$case_id"
+############
+# Internal #
+############
 
-proc __it (case_id ; ; ; block) {
-  # This uses a clean directory
-  echo TODO
-}
-
-# is this accessible to users?
-# It can contain a global list of things to run
+var _testing_in_progress = false
+var _test_suite_depth = 0
+var _test_suite_stack = []
+var _tests = {}
 
-# Naming convention: a proc named 'describe' mutates a global named _describe?
-# Or maybe _describe_list ?
+var _num_test_fail = 0
+var _num_test_succ = 0
+var _test_use_color = true
 
-var _describe_list = []
+var _assertion_result = 0 # 0: no assert, 1: failed, 2: errored
 
-proc describe (test_id ; ; ; block) {
-  echo describe
-  #= desc
+proc _start-test-suite (; name) {
+    call _test_suite_stack->append(name)
+}
 
-  # TODO:
-  # - need append
-  # - need ::
-  # _ _describe->append(cmd)
-  #
-  # Need to clean this up
-  # append (_describe, cmd)  # does NOT work!
+proc _end-test-suite {
+    call _test_suite_stack->pop()
+}
 
-  call _describe_list->append(block)
+proc _test-print-indented (msg) {
+    for _ in (0 .. _test_suite_depth) {
+        printf "    "
+    }
+    echo "$msg"
 }
 
-proc Args {
-  echo TODO
+proc _run-test-suite (; suite) {
+    for name, elem in (suite) {
+        if (type (elem) === "Dict") {
+            # It's another suite.
+            var begin = _cyan("begin")
+            _test-print-indented "$begin $name"
+            setglobal _test_suite_depth += 1;
+            _run-test-suite (elem)
+            setglobal _test_suite_depth -= 1;
+            var end = _cyan("end")
+            _test-print-indented "$end"
+        } else {
+            # It's a test case.
+            _run-test-case (name, elem)
+        }
+    }
 }
 
-# Problem: this creates a global variable?
-Args (&spec) {
-  flag --filter 'Regex of test descriptions'
+proc _run-test-case (; name, block) {
+    var test = _yellow("test")
+    for _ in (0 .. _test_suite_depth) {
+        printf "    "
+    }
+    printf "$test $name ..."
+    setglobal _assertion_result = 0
+    try {
+        eval (block)
+    }
+    if (_status === 0) {
+        setglobal _num_test_succ += 1
+        var ok = _green(" ok")
+        printf "$ok"
+        printf '\n'
+    } else {
+        setglobal _num_test_fail += 1
+        printf '\n'
+        if (_assertion_result === 1) {
+            # An assertion failed.
+            var fail = _red("assertion FAILED")
+            _test-print-indented "    $fail"
+        } elif (_assertion_result === 2) {
+            # There was an exception while evaluating an assertion.
+            var exn_in_assert = _red("assertion ERRORED:")
+            _test-print-indented "    $exn_in_assert $_status"
+        } else {
+            # There was an exception in the test case outside of any `assert`.
+            var exn = _red("ERROR:")
+            _test-print-indented "    $exn $_status"
+        }
+    }
 }
 
-proc run-tests {
-  var opt, i = parseArgs(spec, ARGV)
+##########
+# Public #
+##########
 
-  # TODO:
-  # - parse --filter foo, which you can use eggex for!
+proc test-suite (name ; ; ; block) {
+    _start-test-suite (name)
+    eval (block)
+    _end-test-suite
+}
 
-  for cmd in (_describe) {
-    # TODO: print filename and 'describe' name?
+proc test-case (name ; ; ; block) {
+    var test_suite = _tests
+    for suite_name in (_test_suite_stack) {
+        if (not (suite_name in test_suite)) {
+            setvar test_suite[suite_name] = {}
+        }
+        setvar test_suite = test_suite[suite_name]
+    }
+    setvar test_suite[name] = block
+}
+
+proc assert ( ; cond LAZY ) {
+    var success
     try {
-      eval (cmd)
+        setvar success = evalExpr(cond)
     }
     if (_status !== 0) {
-      echo 'failed'
+        setglobal _assertion_result = 2
+        error "exception in assertion"
+    } elif (not success) {
+        setglobal _assertion_result = 1
+        error "assertion failed"
+    }
+}
+
+proc run-tests {
+    if (_testing_in_progress) {
+        error "Cannot run tests while testing is already in progress"
+    }
+    setglobal _testing_in_progress = true
+    setglobal _num_test_fail = 0
+    setglobal _num_test_succ = 0
+
+    _run-test-suite (_tests)
+    setglobal _tests = {}
+    setglobal _testing_in_progress = false
+
+    var total = _num_test_fail + _num_test_succ
+    if (total === 0) {
+        var na = _yellow("0 tests ran")
+        printf "$na"
+        printf '\n'
+    } elif (_num_test_fail === 0) {
+        var success = _green("$total tests succeeded")
+        printf "$success"
+        printf '\n'
+    } else {
+        var failure = _red("$_num_test_fail / $total tests failed")
+        printf "$failure"
+        printf '\n'
     }
-  }
 }