calibration-diff

echo "=== calibration-diff preprocessing ==="
DOCS_DIR="$(cd "${CLAUDE_SKILL_DIR}/../../docs" 2>/dev/null && pwd || echo UNKNOWN)"
BUNDLES_DIR="$(cd "${CLAUDE_SKILL_DIR}/.." 2>/dev/null && pwd || echo UNKNOWN)"
echo "DOCS_DIR=$DOCS_DIR"
echo "BUNDLES_DIR=$BUNDLES_DIR"
echo "PROJECT_DIR=${CLAUDE_PROJECT_DIR:-$(pwd)}"
echo "NOW_ISO=$(date -u +%Y-%m-%dT%H:%M:%SZ)"
echo "GIT_HEAD=$(git rev-parse HEAD 2>/dev/null || echo not-a-git-repo)"
if [ -f .claude/calibration/current ]; then
  CURRENT="$(cat .claude/calibration/current)"
  echo "CURRENT_RUN=$CURRENT"
else
  echo "CURRENT_RUN=(none)"
fi
echo "--- recent calibration runs (newest first) ---"
ls -1dt .claude/calibration/*/ 2>/dev/null | head -5 || echo "(none)"
echo "=== end preprocessing ==="

calibration-diff — re-evaluate against the last baseline

You are the diff-only flow. You run the calibration evaluator pass 2 against the previous run's baseline reports and stop. No planner, no calibrator, no edits.

What to do

1. Resolve DOCS_DIR, BUNDLES_DIR, PROJECT_DIR from the preprocessing block.
2. Pick the run to diff against. Default: the latest run folder with a complete baseline_reports list in plan.md (read it; the field is set when the evaluator finishes Pass 1). If the user passed a run identifier in $ARGUMENTS (e.g. a timestamp), use that one. If no run is found, print one line: No prior calibration run found — run /calibrate first to establish a baseline. and stop. Also read the picked run's plugin_filter field — the delta must re-audit the same plugin scope as the baseline, so reuse it verbatim (don't take a new --plugins flag here).
3. Spawn the evaluator (Pass 2):

   Agent(calibration-evaluator)
   Pass: 2 (delta).
   Run folder: <abs of the picked run>.
   Plan: <run>/plan.md.
   Baseline reports: <the names from plan.md baseline_reports>.
   Rubric dir: <DOCS_DIR>.
   Bundles dir: <BUNDLES_DIR>.
   Project dir: <PROJECT_DIR>.
   Plugin filter: <the picked run's plugin_filter from plan.md, or empty = all plugins>.
   

   The evaluator will write eval-delta-<ts>.md into the run folder and update plan.md's last_evaluation field.
4. Print a short summary. Read plan.md's now-updated last_evaluation field; print:
   • Diffing against: <run folder> (baseline taken <plan.started>).
   • Before: C<n> H<n> M<n> L<n>. After: C<n> H<n> M<n> L<n>.
   • Resolved <n> · partial <n> · open <n> · new <n>.
   • Delta report: <run>/eval-delta-<ts>.md.
   • → Run /calibrate to plan fixes for the open + new findings.