polish docs

docs/runner_reference.md

@@ -65,7 +65,7 @@ The download cache functionality must be enabled in the configuration. The relev
 Details:
 
 * Enabling the cache changes **where downloaded files are stored**: if the cache is enabled, they're stored in the cache directory; otherwise, they're stored under the triggering run directory.
-* The cache is **keyed by URI**: when a workflow starts with a URI file input, a cached file is used if previously stored for the same URI. This doesn't depend on which task/workflow is running, and doesn't use checksums or timestamps of the file contents. Therefore, the cache should only be used with immutable remote files, or if there's no need for immediate coherence with remote content changes.
+* The cache is **keyed by URI**: when a workflow starts with a URI file input, a cached file is used if previously stored for the same URI. This doesn't depend on which task/workflow is running, and doesn't use checksums or timestamps of the file contents. (Therefore, the cache should only be used with immutable remote files, or if there's no need for immediate coherence with remote content changes.)
 * If needed, the `miniwdl localize` subcommand can **"prime" the local cache** with URIs found in a given JSON input template (or a simple list of URIs) before actually running any workflow.
 * Cached files that are no longer needed can simply be **deleted from the cache directory**, once they're no longer in use by a running workflow.
-* Miniwdl itself doesn't delete files from the cache, but to support a **cleanup process**, it updates the access timestamp (atime) and opens a shared `flock()` on any cached file it's using. The script [examples/clean_download_cache.sh](https://github.com/chanzuckerberg/miniwdl/blob/master/examples/clean_download_cache.sh) illustrates a process to shrink the cache to a desired maximum size, by evicting the least-recently used files that can be exclusively flocked (the latter condition needed only if the cleaner must run alongside concurrent workflows).
+* Miniwdl itself doesn't delete files from the cache, but to support an **external cleanup process**, it updates the access timestamp (atime) and opens a shared `flock()` on any cached file it's using. The script [examples/clean_download_cache.sh](https://github.com/chanzuckerberg/miniwdl/blob/master/examples/clean_download_cache.sh) illustrates a process to shrink the cache to a desired maximum size, by evicting the least-recently used files that can be exclusively flocked (the latter condition needed only if the cleaner must run alongside concurrent workflows).