--- a/taskcluster/taskgraph/transforms/job/toolchain.py
+++ b/taskcluster/taskgraph/transforms/job/toolchain.py
@@ -24,19 +24,24 @@ from taskgraph.util.hash import hash_pat
 from taskgraph import GECKO
 
 
 TOOLCHAIN_INDEX = 'gecko.cache.level-{level}.toolchains.v1.{name}.{digest}'
 
 toolchain_run_schema = Schema({
     Required('using'): 'toolchain-script',
 
-    # the script (in taskcluster/scripts/misc) to run
+    # The script (in taskcluster/scripts/misc) to run.
+    # Python scripts are invoked with `mach python` so vendored libraries
+    # are available.
     Required('script'): basestring,
 
+    # Arguments to pass to the script.
+    Optional('arguments'): basestring,
+
     # If not false, tooltool downloads will be enabled via relengAPIProxy
     # for either just public files, or all files.  Not supported on Windows
     Required('tooltool-downloads', default=False): Any(
         False,
         'public',
         'internal',
     ),
 
@@ -59,31 +64,36 @@ def add_optimizations(config, run, taskd
     files.append('taskcluster/taskgraph/transforms/job/toolchain.py')
     # The script
     files.append('taskcluster/scripts/misc/{}'.format(run['script']))
     # Tooltool manifest if any is defined:
     tooltool_manifest = taskdesc['worker']['env'].get('TOOLTOOL_MANIFEST')
     if tooltool_manifest:
         files.append(tooltool_manifest)
 
-    digest = hash_paths(GECKO, files)
+    # Accumulate dependency hashes for index generation.
+    data = [hash_paths(GECKO, files)]
 
     # If the task has dependencies, we need those dependencies to influence
     # the index path. So take the digest from the files above, add the list
     # of its dependencies, and hash the aggregate.
     # If the task has no dependencies, just use the digest from above.
     deps = taskdesc['dependencies']
     if deps:
-        data = [digest] + sorted(deps.values())
-        digest = hashlib.sha256('\n'.join(data)).hexdigest()
+        data.extend(sorted(deps.values()))
+
+    # Likewise script arguments should influence the index.
+    args = run.get('arguments')
+    if args:
+        data.append(args)
 
     label = taskdesc['label']
     subs = {
         'name': label.replace('%s-' % config.kind, ''),
-        'digest': digest,
+        'digest': hashlib.sha256('\n'.join(data)).hexdigest()
     }
 
     optimizations = taskdesc.setdefault('optimizations', [])
 
     # We'll try to find a cached version of the toolchain at levels above
     # and including the current level, starting at the highest level.
     for level in reversed(range(int(config.params['level']), 4)):
         subs['level'] = level