ashlr-benchmark

Benchmark the plugin's snipCompact savings on the user's actual code.

Steps:

1. Determine a target directory:

   • Prefer <cwd>/src if it exists.

   • Otherwise try <cwd>/lib, <cwd>/app, then <cwd> itself.

   • If none of these contain any .ts, .tsx, .js, .jsx, .py, .go, .rs, .java, or similar source files, tell the user:

     ashlr-benchmark: no source files detected in <cwd>.
     Run it against a specific file instead, e.g.:
       bun run ${CLAUDE_PLUGIN_ROOT}/servers/bench.ts --path <file>
     

     and stop.

2. Run via Bash. The plugin root is exposed to hooks and commands via the ${CLAUDE_PLUGIN_ROOT} environment variable. If that variable is empty or unset, the plugin is not loaded — abort with this exact message and stop:

   ashlr plugin not loaded — run /plugin install ashlr@ashlr-marketplace first.
   

   Otherwise:

   bun run ${CLAUDE_PLUGIN_ROOT}/servers/bench.ts --dir <target-dir>
   

3. Parse the output. The bench script prints a table of files with their raw byte counts vs. snipCompact byte counts and a percent saved. Reproduce the table in the response, then prepend a one-line headline:

   mean savings on files ≥ 2KB: X%
   

   Compute the mean across only files whose raw size is ≥ 2048 bytes (smaller files are pass-through in the redirect hook and skew the average).

4. If the bench script fails (missing file, exit non-zero), surface stderr verbatim and suggest running with --path <single-file> to narrow the problem.

No preamble, no trailing summary — just the headline plus the table.