[guyoron1]

Summary

Closes #2653.

condense_unified_diff prints ALL change lines in full (honoring the "Never truncate diff content" comment), then appends ... +N more when total changes exceed 10. This misleads LLM consumers into requesting "remaining" data that was already shown, wasting tokens on redundant fetches.

Before: 200 changes printed, then ... +190 more (implies 190 lines were hidden)
After: 200 changes printed, no misleading indicator

Test plan

• Updated test_condense_unified_diff_overflow_count_accuracy → renamed to test_condense_unified_diff_large_no_false_overflow_indicator, now asserts NO overflow message appears and verifies all lines are present
• Existing test_condense_unified_diff_no_false_overflow continues to pass
• All 23 diff_cmd tests pass
• cargo fmt --all && cargo clippy --all-targets && cargo test --all clean (2237 tests, 0 failures)

[KuSh]

Approving. Reviewed at depth and found nothing to change in the fix itself.

What I verified:

• The changes vec is never capped — every +/- line is pushed unconditionally and all are emitted — so total > 10 → "+{total-10} more" was pure fabrication, always claiming hidden lines that had in fact just been printed. Removing it is the correct fix, not a masking of real truncation.
• added/removed are still consumed by the [file] {} (+{} -{}) header, so no dead variables.
• condense_unified_diff has one caller (run_stdin), and no other module, test or doc references the overflow string. The other ... +N more occurrences in the repo (deps.rs, pip_cmd.rs, summary.rs, prettier_cmd.rs) back real truncation and are untouched.
• Output only shrinks, so never_worse's raw-fallback behaviour is unaffected.
• cargo fmt --check, cargo clippy --all-targets and cargo test --all (2583 unit + integration) are clean on the head.

Confirmed live on 0.46.0 before the change, so this is a real user-facing lie, not a theoretical one:

$ rtk diff - < twelve_changes.diff | tail -1
  ... +14 more          # after printing all 24 change lines

I pushed one commit to your branch (7aa6347): the test module's section header still read // --- truncation accuracy --- and test_condense_unified_diff_no_false_overflow still mentioned a "10-line display cap". Neither describes the code once the indicator is gone, and leaving them would have re-planted the same misconception this PR removes. Shout if you'd rather word them differently.

Thanks for this one — it's a small diff, but "the filter tells the model that data is missing when it isn't" costs a re-fetch every time, and that's the expensive kind of wrong.