/compress

The user wants to compress content using the token-optimizer plugin.

Argument: $ARGUMENTS

If the argument looks like a file path (starts with / or ./ or ~), use Read to read it. Otherwise, treat the argument itself as the content to compress.

Pick the mode by content type (same rules as the skill):

• Code the user wants an overview of → add --ast (drops function bodies, reversible via the retrieve tool). Skip --ast if they need to run/debug it. This is the clearest win.
• JSON / structured data → default is already strong; --aggressive is safe to add but may only help a little (the engine auto-tunes).
• Prose / natural language → no flag (aggressive is lossy on prose).

Then call the compressor (self-contained dist/ bundle; add the chosen flag):

!echo "{CONTENT}" | node "${CLAUDE_PLUGIN_ROOT:-$HOME/.claude/plugins/agentone-token-compression}/dist/scripts/compress-stdin.js" --aggressive

(Replace {CONTENT} with the actual content via stdin; the script reads from stdin. Drop --aggressive for prose; use --ast instead for code overviews.)

Show the user:

1. Original: byte count, estimated token count, detected content type
2. Compressed: byte count, estimated token count, % saved
3. Technique used: which optimizers fired (dedupe, dictionary, JSON minify, AST body-drop, etc.)
4. Compressed text (in a code block)

Report the actual savedPct — don't promise a number in advance. Savings vary a lot by content: JSON ~50%, repetitive code 80%+, but logs with unique per-line values (ids/timestamps) may compress very little (~0%).

If the content contains API keys or other secrets, point out that those were preserved verbatim by the entropy/secret detectors.