/compact [instructions] does not actually trigger compaction

Bug Description

/compact [instructions] does not perform compaction. Running /compact with any arguments returns instantly and does not reduce context usage. Running bare /compact (no arguments) works correctly — it takes visible processing time and context usage drops as expected.

Steps to Reproduce

1. Start a session and do enough work to build up context
2. Run /context to note current usage
3. Run /compact preserve the key decisions and current task
4. Observe: the command returns instantly (no processing time)
5. Run /context again — usage has not changed
6. Now run bare /compact (no arguments)
7. Observe: this takes several seconds to process
8. Run /context — usage has dropped as expected

Expected Behavior

/compact [instructions] should trigger compaction with the instructions guiding what the summary preserves, per the documentation:

/compact [instructions] — Compact conversation with optional focus instructions

Actual Behavior

/compact [instructions] appears to be parsed as a regular prompt to the model rather than triggering the compaction system. No compaction occurs — context usage is unchanged.

Only bare /compact (no arguments) triggers actual compaction.

Environment

• Platform: macOS (Darwin 25.2.0)
• Model: Claude Opus 4.6
• Plan: Max subscription

Impact

The documented feature of providing focus instructions to guide compaction does not work. Users relying on /compact [instructions] to preserve specific context during compaction are getting no compaction at all, leading to eventual context exhaustion.

This also affects any skills or workflows built around composing targeted /compact instructions (e.g., the community pattern of auditing context and generating a focused compact command).