grep: per-file match cap causes LLM retries (10 matches hardcoded)

[FlorianBruniaux]

Problem

rtk grep caps results at 10 matches per file (src/grep_cmd.rs:113 — .take(10)), with a global max of 50.

This produces a "+N more" truncation that tells Claude signal exists but withholds it. Claude then re-runs the search with different args to recover the missing results — burning more tokens than a raw grep would have.

Root cause

src/grep_cmd.rs:113:

.take(10)  // hardcoded per-file cap

src/main.rs:309: global max = 50.

Impact

• LLM sees "+N more" and loops to find the missing matches
• Each retry costs more tokens than the original search
• Net result: RTK makes token consumption worse than raw grep

Proposed fix

• Remove per-file cap (or raise to 50)
• Raise global max from 50 → 100
• Add optional --max-per-file <N> flag for users who want stricter limits

Acceptance criteria

• rtk grep "pattern" returns all matches per file up to the global max
• No "+N more" truncation that signals hidden results
• Token savings still ≥50% vs raw grep (from deduplication and grouping, not caps)

[pszymkowiak]

Fixed in v0.30.0 — grep per-file match cap raised from 10 to 25. Verified with release binary.