fix-findingslisted

# Fix Findings !`PROJECT_DIR="${CLAUDE_PROJECT_DIR:-$(git rev-parse --show-toplevel)}" && mkdir -p "$PROJECT_DIR/.claude" && touch "$PROJECT_DIR/.claude/.fix-findings-active" && echo "fix-findings marker created at $PROJECT_DIR/.claude/.fix-findings-active — post-review edit lock open for this run"` Take the review findings a human has already chosen, have a **fresh** sub-agent write each fix, have a **second** fresh sub-agent try to break it, and report both back. Hand off to a `/pre-merge` re-run, which is the only thing that re-stamps. ## Why this skill exists `/pre-merge` delegates the review to a sub-agent that did not write the code, because the session that wrote it has already demonstrated it cannot see these particular defects. Then the human picks findings, and the pipeline hands the *fix* back to that same blind session. The responsive commit is the most defect-dense code on the branch and the one commit with no independent reader. Cohen's rule is that an author's job is to *annotate for* a reviewer, not to be one. This skill applies it one link later: the fix's author must not be its only reader. Leveson's is that a controller may not treat "sent" as "executed" without feedback — a fixer's own green test run is exactly that unearned inference, which is why a second agent attacks the fix rather than the fixer re-checking it. ## Invocation Position **Invoked helper, and user-invoked only.** It fires when a human types `/fix-findings <numbers>`; it never self-