diff --git a/test/minimal-command.wasm b/test/minimal-command.wasm
new file mode 100644
index 00000000..ca050857
Binary files /dev/null and b/test/minimal-command.wasm differ
diff --git a/test/sources/generate-wasm.sh b/test/sources/generate-wasm.sh
new file mode 100755
index 00000000..9d5c5fa1
--- /dev/null
+++ b/test/sources/generate-wasm.sh
@@ -0,0 +1,8 @@
+#!/bin/bash
+set -ueo pipefail
+
+for wat in wat/*.wat; do
+    wat2wasm "$wat"
+done
+
+mv *.wasm ..
diff --git a/test/sources/wat/minimal-command.wat b/test/sources/wat/minimal-command.wat
new file mode 100644
index 00000000..6e9f4f5f
--- /dev/null
+++ b/test/sources/wat/minimal-command.wat
@@ -0,0 +1,3 @@
+(module
+    (func (export "_start"))
+)
diff --git a/tools/test-harness/exit_status_zero b/tools/test-harness/exit_status_zero
new file mode 100644
index 00000000..573541ac
--- /dev/null
+++ b/tools/test-harness/exit_status_zero
@@ -0,0 +1 @@
+0
diff --git a/tools/test-harness/run.sh b/tools/test-harness/run.sh
new file mode 100755
index 00000000..218856a2
--- /dev/null
+++ b/tools/test-harness/run.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+set -ueo pipefail
+
+# Top-level test runner. Usage is "run.sh" to run tests in compile-only mode,
+# or "run.sh <runwasi>" where <runwasi> is a WASI-capable runtime to run the
+# tests in full compile and execute mode.
+
+# Determine the wasm runtime to use.
+runwasi="$1"
+
+tooldir=$(dirname $0)
+
+echo "===== Testing ====="
+for file in *.wasm; do
+    "$tooldir/testcase.sh" "$runwasi" "$file"
+done
+cd - >/dev/null
diff --git a/tools/test-harness/testcase.sh b/tools/test-harness/testcase.sh
new file mode 100755
index 00000000..482fdddc
--- /dev/null
+++ b/tools/test-harness/testcase.sh
@@ -0,0 +1,90 @@
+#!/bin/bash
+set -ueo pipefail
+
+# A simple testcase runner that runs a command, captures all its command-line
+# outputs, and compares them against expected outputs.
+
+# Command-line parsing; this script is meant to be run from a higher-level
+# script, so don't do anything fancy.
+runwasi="$1"
+input="$2"
+
+# Compile names for generated files.
+wasm="$input"
+stdout_observed="$input.stdout.observed"
+stderr_observed="$input.stderr.observed"
+exit_status_observed="$input.exit_status.observed"
+tooldir=$(dirname $0)
+
+echo "Testing $input..."
+
+# Determine the input file to write to stdin.
+if [ -e "$input.stdin" ]; then
+  stdin="$input.stdin"
+else
+  stdin="/dev/null"
+fi
+
+# Determine any environment variables to set.
+if [ -e "$input.env" ]; then
+    env=$(sed -e 's/^/--env /' < "$input.env")
+else
+    env=""
+fi
+
+# Determine a preopened directory to provide.
+if [ -e "$input.dir" ]; then
+    dir="--dir $input.dir"
+    dirarg="$input.dir"
+else
+    dir=""
+    dirarg=""
+fi
+
+# Run the test, capturing stdout, stderr, and the exit status.
+exit_status=0
+"$runwasi" $env $dir "$wasm" $dirarg \
+    < "$stdin" \
+    > "$stdout_observed" \
+    2> "$stderr_observed" \
+    || exit_status=$?
+echo $exit_status > "$exit_status_observed"
+
+# Determine the reference files to compare with.
+if [ -e "$input.stdout.expected" ]; then
+  stdout_expected="$input.stdout.expected"
+
+  # Apply output filters.
+  if [ -e "$input.stdout.expected.filter" ]; then
+      cat "$stdout_observed" \
+          | "$input.stdout.expected.filter" \
+          > "${stdout_observed}.filtered"
+      stdout_observed="${stdout_observed}.filtered"
+  fi
+else
+  stdout_expected="/dev/null"
+fi
+if [ -e "$input.stderr.expected" ]; then
+  stderr_expected="$input.stderr.expected"
+
+  # Apply output filters.
+  if [ -e "$input.stderr.expected.filter" ]; then
+      cat "$stderr_observed" \
+          | "./$input.stderr.expected.filter" \
+          > "${stderr_observed}.filtered"
+      stderr_observed="${stderr_observed}.filtered"
+  fi
+else
+  stderr_expected="/dev/null"
+fi
+if [ -e "$input.exit_status.expected" ]; then
+  exit_status_expected="$tooldir/$input.exit_status.expected"
+else
+  exit_status_expected="$tooldir/exit_status_zero"
+fi
+
+# If there are any differences, diff will return a non-zero exit status, and
+# since this script uses "set -e", it will return a non-zero exit status too.
+diff -u "$stderr_expected" "$stderr_observed"
+diff -u "$stdout_expected" "$stdout_observed"
+diff -u "$exit_status_expected" "$exit_status_observed"