evaluate

Step 1: Resolve evaluation target

1. If the user specifies a path after @evaluate (e.g., @evaluate worktrees/omk-foo), use that path.
2. Otherwise, check .active-submodule:

if [ -f .active-submodule ]; then
  jq -r '.worktree // empty' .active-submodule
fi

1. If a worktree path is found, use it. Otherwise, use project root.

Set the resolved path as EVAL_DIR.

Step 2: Gather context

cd "$EVAL_DIR"
PLAN=$(cat docs/plans/*.md 2>/dev/null | head -200)
DIFF=$(git diff --stat && git diff)
[ -z "$DIFF" ] && DIFF=$(git diff --cached --stat && git diff --cached)

Step 3: Dispatch 4 parallel evaluator subagents

Dispatch ALL 4 subagents in parallel (one use_subagent call with 4 entries). Each subagent receives the plan goals, diff, and its specific evaluation mandate.

Every subagent MUST:

• Fill ALL mandatory tables — empty or missing table rows = REJECTED, re-dispatch
• Answer its Canary question (proves source code was actually read)
• Classify each finding: CRITICAL / HIGH / MEDIUM / LOW
• End with exactly: Verdict: PASS or Verdict: FAIL
• Missing verdict = malformed → REJECTED, re-dispatch

---

Subagent #1: "Refactoring Expert" — Simplicity + Maintainability

Persona: A senior engineer who believes the best code is code that doesn't exist. Your job is to find things to delete or simplify.

Read all modified files in EVAL_DIR. Then fill EVERY table below — empty table = REJECTED.

Table A — Long functions (>50 lines):

Function	File:Line	Lines	Can split?	Split plan	Reason if not

If no functions >50 lines, write one row: "None found — all functions ≤50 lines."

Table B — Exception handling:

Location (file:line)	Catches what	Necessary?	What if removed

If no try/except blocks, write one row: "No exception handlers found."

Table C — Abstraction layers: