diff --git a/.githooks/pre-push b/.githooks/pre-push
new file mode 100755
index 00000000..34393dae
--- /dev/null
+++ b/.githooks/pre-push
@@ -0,0 +1,30 @@
+#!/usr/bin/bash
+set -euo pipefail
+# Git pre-push hook — runs skill eval for changed skills (non-blocking)
+#
+# Install shared hooks from the repo root: make setup-githooks
+# Enable eval on push: export RUN_EVAL=1
+
+if [[ "${RUN_EVAL:-0}" == "1" ]]; then
+    REPO_ROOT="$(git rev-parse --show-toplevel)"
+    EVAL_SHA="$(git rev-parse HEAD)"
+
+    # Extract owner/repo from origin remote (SSH or HTTPS)
+    EVAL_REPO="$(git remote get-url origin \
+        | sed -E 's#^(https?://github\.com/|git@github\.com:)##' \
+        | sed 's/\.git$//')"
+
+    EVAL_LOG="/tmp/skill-eval-${EVAL_SHA:0:8}.log"
+
+    echo "Skill eval will run in the background (log: ${EVAL_LOG})"
+
+    (
+        nohup env \
+            EVAL_SHA="${EVAL_SHA}" \
+            EVAL_REPO="${EVAL_REPO}" \
+            "${REPO_ROOT}/scripts/eval-skill.sh" \
+            > "${EVAL_LOG}" 2>&1
+    ) &
+    disown
+fi
+exit 0
diff --git a/.gitignore b/.gitignore
index 1524e53b..309454d6 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,4 +1,5 @@
 .reports
+eval/runs/
 .mcp.json
 !plugins/**/.mcp.json
 .DS_Store
diff --git a/README.md b/README.md
index 8c97ddc0..e817fcab 100644
--- a/README.md
+++ b/README.md
@@ -24,7 +24,10 @@ cd edge-tooling
 make setup-githooks
 ```
 
-`make setup-githooks` installs a pre-commit hook that runs markdownlint on staged `.md` files.
+`make setup-githooks` installs shared git hooks:
+
+- **pre-commit** — runs markdownlint on staged `.md` files
+- **pre-push** — optionally evaluates changed skills via the agent-eval-harness (`RUN_EVAL=1 git push`)
 
 Prerequisites: Node.js (for markdownlint), Python 3, Bash.
 
diff --git a/scripts/eval-skill.sh b/scripts/eval-skill.sh
new file mode 100755
index 00000000..02ecf782
--- /dev/null
+++ b/scripts/eval-skill.sh
@@ -0,0 +1,171 @@
+#!/usr/bin/bash
+set -euo pipefail
+
+EVAL_MODEL="${EVAL_MODEL:-claude-sonnet-4-6}"
+BASE_BRANCH="${BASE_BRANCH:-origin/main}"
+EVAL_WORKDIR=$(mktemp -d -t skill-eval-XXXXXX)
+EVAL_RUNS_DIR="${AGENT_EVAL_RUNS_DIR:-eval/runs}"
+
+trap 'rm -rf "${EVAL_WORKDIR}"' EXIT
+
+# ── Commit status helper ───────────────────────────────────
+# Posts GitHub commit status when EVAL_SHA and EVAL_REPO are
+# set (by the pre-push hook). No-op on manual invocation.
+post_status() {
+    local state="$1"
+    local description="$2"
+
+    [[ -z "${EVAL_SHA:-}" || -z "${EVAL_REPO:-}" ]] && return 0
+    command -v gh >/dev/null 2>&1 || return 0
+
+    gh api "repos/${EVAL_REPO}/statuses/${EVAL_SHA}" \
+        -f state="${state}" \
+        -f context="skill-eval" \
+        -f description="${description}" \
+        2>/dev/null || true
+}
+
+# ── PR comment helper ──────────────────────────────────────
+# Posts eval report as a PR comment on the current branch's PR.
+post_pr_comment() {
+    local report_file="$1"
+
+    [[ -z "${EVAL_SHA:-}" || -z "${EVAL_REPO:-}" ]] && return 0
+    [[ ! -f "${report_file}" ]] && return 0
+    command -v gh >/dev/null 2>&1 || return 0
+
+    local body
+    body="## Skill Eval Report
+
+$(cat "${report_file}")"
+
+    gh pr comment --edit-last --body "${body}" 2>/dev/null \
+        || gh pr comment --body "${body}" 2>/dev/null || true
+}
+
+echo "╔══════════════════════════════════════════════╗"
+echo "║  Skill Eval — scanning for changed skills    ║"
+echo "╚══════════════════════════════════════════════╝"
+echo ""
+
+# ── Step 1: find changed SKILL.md files ─────────────────────
+# Scope to the current user's commits only — other people's merged
+# work on the branch should not trigger eval.
+GIT_USER="$(git config user.name)"
+mapfile -t CHANGED_SKILLS < <(
+    git log --author="${GIT_USER}" --format="%H" "${BASE_BRANCH}"..HEAD 2>/dev/null \
+    | xargs -rI{} git diff-tree --no-commit-id -r --name-only {} \
+    | grep '/skills/.*SKILL\.md$' \
+    | sort -u \
+    || true)
+
+if [[ ${#CHANGED_SKILLS[@]} -eq 0 ]]; then
+    echo "No skill files changed, nothing to evaluate."
+    exit 0
+fi
+
+echo "Changed skills:"
+printf '  • %s\n' "${CHANGED_SKILLS[@]}"
+echo ""
+
+post_status "pending" "Evaluating ${#CHANGED_SKILLS[@]} skill(s)..."
+
+EVAL_STATUS_POSTED=0
+trap 'if [[ "${EVAL_STATUS_POSTED}" -eq 0 ]]; then post_status "error" "Eval crashed unexpectedly"; fi; rm -rf "${EVAL_WORKDIR}"' EXIT
+
+# ── Step 2: for each changed skill, analyze → generate → run ─
+TOTAL=0
+PASSED=0
+FAILED=0
+REPORT_FILE="${EVAL_WORKDIR}/summary.txt"
+LAST_ANALYSIS=""
+
+for skill_file in "${CHANGED_SKILLS[@]}"; do
+    TOTAL=$((TOTAL + 1))
+
+    # Extract plugin and skill name from path
+    # e.g. plugins/two-node/skills/cluster-diagnostic/SKILL.md
+    plugin=$(echo "${skill_file}" | cut -d'/' -f2)
+    skill_dir=$(echo "${skill_file}" | rev | cut -d'/' -f2 | rev)
+    skill_name="${plugin}:${skill_dir}"
+
+    config="${EVAL_WORKDIR}/${plugin}-${skill_dir}.yaml"
+    cases_dir="${EVAL_WORKDIR}/${plugin}-${skill_dir}/cases"
+    mkdir -p "${cases_dir}"
+
+    echo "════════════════════════════════════════════"
+    echo "Evaluating: ${skill_name}"
+    echo "  Skill file: ${skill_file}"
+    echo "  Temp config: ${config}"
+    echo "════════════════════════════════════════════"
+
+    # Step 2a: Analyze — generate eval config with judges
+    echo "[1/3] Analyzing skill..."
+    if ! claude -p "/eval-analyze --skill ${skill_name} --config ${config}" 2>&1; then
+        echo "  ⚠ eval-analyze failed for ${skill_name}, skipping."
+        FAILED=$((FAILED + 1))
+        echo "SKIP  ${skill_name}  (eval-analyze failed)" >> "${REPORT_FILE}"
+        continue
+    fi
+
+    # Step 2b: Generate — create test scenarios
+    echo "[2/3] Generating test scenarios..."
+    if ! claude -p "/eval-dataset --config ${config}" 2>&1; then
+        echo "  ⚠ eval-dataset failed for ${skill_name}, skipping."
+        FAILED=$((FAILED + 1))
+        echo "SKIP  ${skill_name}  (eval-dataset failed)" >> "${REPORT_FILE}"
+        continue
+    fi
+
+    # Step 2c: Run — execute eval and score
+    echo "[3/3] Running evaluation..."
+    if claude -p "/eval-run --model ${EVAL_MODEL} --config ${config}" 2>&1; then
+        PASSED=$((PASSED + 1))
+        echo "PASS  ${skill_name}" >> "${REPORT_FILE}"
+    else
+        FAILED=$((FAILED + 1))
+        echo "FAIL  ${skill_name}" >> "${REPORT_FILE}"
+    fi
+
+    # Find the most recent analysis.md for this run
+    latest_run=$(find "${EVAL_RUNS_DIR}" -name "analysis.md" -newer "${config}" 2>/dev/null \
+        | sort | tail -1)
+    if [[ -n "${latest_run}" ]]; then
+        LAST_ANALYSIS="${latest_run}"
+    fi
+
+    echo ""
+done
+
+# ── Step 3: print summary report ────────────────────────────
+echo ""
+echo "╔══════════════════════════════════════════════╗"
+echo "║  Skill Eval Report                           ║"
+echo "╠══════════════════════════════════════════════╣"
+printf "║  Skills evaluated: %-24s ║\n" "${TOTAL}"
+printf "║  Passed:           %-24s ║\n" "${PASSED}"
+printf "║  Failed/Skipped:   %-24s ║\n" "${FAILED}"
+echo "╠══════════════════════════════════════════════╣"
+
+if [[ -f "${REPORT_FILE}" ]]; then
+    while IFS= read -r line; do
+        printf "║  %-43s ║\n" "${line}"
+    done < "${REPORT_FILE}"
+fi
+echo "╚══════════════════════════════════════════════╝"
+
+# ── Step 4: post results to GitHub ──────────────────────────
+if [[ "${FAILED}" -gt 0 ]]; then
+    post_status "failure" "${PASSED}/${TOTAL} skills passed, ${FAILED} failed"
+else
+    post_status "success" "${PASSED}/${TOTAL} skills passed"
+fi
+
+if [[ -n "${LAST_ANALYSIS}" ]]; then
+    post_pr_comment "${LAST_ANALYSIS}"
+fi
+
+EVAL_STATUS_POSTED=1
+
+# Always exit 0 — non-blocking
+exit 0