Hook: chained commands (cd dir && cmd) are never rewritten

[Rmohid]

Problem

Claude Code frequently emits chained commands like:

cd /Users/me/code/project && git status
cd /some/dir && python3 -m pytest tests/ -v
cd /foo && git add . && git commit -m "msg"

The hook script sets FIRST_CMD="$CMD" and then matches against patterns like ^git[[:space:]]+status. Since the full command starts with cd, none of the patterns match, and the hook silently passes through.

rtk discover shows this is the dominant source of missed savings -- 271 git status, 65 tail, 81 grep, 105 ls calls missed in 30 days of usage (~84K tokens saveable).

Fix

Extract the last command segment after && or ; for pattern matching, then reassemble the full command after rewriting:

CHAIN_PREFIX=""
FINAL_CMD="$CMD"

if [[ "$CMD" == *" && "* ]]; then
  CHAIN_PREFIX="${CMD% && *} && "
  FINAL_CMD="${CMD##* && }"
elif [[ "$CMD" == *" ; "* ]]; then
  CHAIN_PREFIX="${CMD% ; *} ; "
  FINAL_CMD="${CMD##* ; }"
fi

# ... match/rewrite against FINAL_CMD ...

# Reassemble
if [ -n "$CHAIN_PREFIX" ]; then
  REWRITTEN="${CHAIN_PREFIX}${REWRITTEN}"
fi

This uses bash parameter expansion (no regex issues with && in =~).

Environment

• rtk 0.15.3
• macOS (Tahoe)
• Claude Code CLI

[ahundt]

I have hooks that do this, it is actually harder than that to do correctly. You have to use an bash lexer to actually parse the commands and determine if a given chunk is actually a command or part of other commands and params. There are also edge cases like heredocs and comments and many more to get it right.

[ahundt]

Hey @Rmohid I've implemented this request in #131 if you can give it a try let me know what you think!

[F0rty-Tw0]

I had the same issue, and just updated my RTK hooks to catch that.

[pszymkowiak]

Tracked in PR #156 (Hook Engine + Chained Command Rewriting).

[Rmohid]

Verified fix: PR #156 adds Rust hook engine with lexer/chain parser. cd /tmp && git status correctly rewrites to cd /tmp && rtk git status. Build passes, 541 tests pass (336 in hook engine, 15 chain-specific). PR #150 also handles this case.