#!/usr/bin/env bash
# Deterministic security checklist seed scanner (paths + diff hunks).
# Hits here are hard-fail regardless of model agreement (see policy.security_agreement).
set -euo pipefail

FILES_LIST="${1:?changed-files list}"
OUT="${2:?out json}"
ROOT="${GATEKEEP_ROOT:-.}"
REPO="${GITHUB_REPOSITORY:-}"
EVENT="${GITHUB_EVENT_PATH:-}"

hits=()

while IFS= read -r f; do
  [[ -z "$f" ]] && continue
  case "$f" in
    .github/workflows/*|.github/actions/*)
      hits+=("CI workflow tampering surface: $f")
      ;;
    *secret*|*credential*|.env*|.env)
      hits+=("Secrets/credential path touched: $f")
      ;;
  esac
done < "$FILES_LIST"

# Diff scan for high-signal patterns (best-effort; forge-agnostic)
if [[ -n "$EVENT" && -f "$EVENT" && -n "$REPO" ]]; then
  PR=$(python3 -c 'import json,sys; print(json.load(open(sys.argv[1]))["pull_request"]["number"])' "$EVENT")
  ACTION_DIR="${GATEKEEP_ACTION:-$(cd "$(dirname "$0")/.." && pwd)}"
  TMP=$(mktemp)
  # Prefer .diff endpoint (works on Gitea + GitHub)
  SERVER="${GITHUB_SERVER_URL:-}"
  TOKEN="${GITHUB_TOKEN:-${GITEA_TOKEN:-}}"
  if [[ -n "$SERVER" ]]; then
    curl -sS -H "Authorization: token ${TOKEN}" \
      "${SERVER%/}/${REPO}/pulls/${PR}.diff" -o "$TMP" 2>/dev/null \
      || curl -sS -H "Authorization: token ${TOKEN}" \
        "${GITHUB_API_URL%/}/repos/${REPO}/pulls/${PR}.diff" -o "$TMP" 2>/dev/null \
      || true
  fi
  if [[ ! -s "$TMP" ]]; then
    python3 "$ACTION_DIR/scripts/forge_api.py" GET "/repos/${REPO}/pulls/${PR}.diff" >"$TMP" 2>/dev/null || true
  fi
  DIFF=$(cat "$TMP" 2>/dev/null || true)
  # Only look at added lines
  ADDED=$(grep -E '^\+' "$TMP" | grep -vE '^\+\+\+' || true)
  check() {
    local name="$1" pat="$2"
    if printf '%s\n' "$ADDED" | grep -Eiq -- "$pat"; then
      hits+=("$name")
    fi
  }
  check "Hardcoded secrets/tokens pattern" '(api[_-]?key|secret|token|password|private[_-]?key)\s*[:=]\s*['\''\"]?[A-Za-z0-9_\-]{16,}'
  check "Prompt injection via repo instruction files" '(ignore (all )?(previous|above) instructions|you are now|system prompt)'
  check "Possible SQL injection concat" '(execute\(|query\(|raw\().*\+|(SELECT|DROP|UNION).*\+'
  check "Possible command injection" '(exec\(|spawn\(|os\.system\(|subprocess\.[a-z]+\([^)]*shell\s*=\s*True)'
  check "Unsafe deserialization" '(pickle\.loads|yaml\.load\(|unserialize\(|ObjectInputStream)'
  rm -f "$TMP"
fi

HARD=false
if ((${#hits[@]} > 0)); then
  HARD=true
fi

python3 - "$OUT" "$HARD" "${hits[@]}" <<'PY'
import json,sys
out=sys.argv[1]
hard=sys.argv[2]=="true"
hits=sys.argv[3:]
json.dump({"hard_fail": hard, "hits": hits, "checklist": [
  "hardcoded secrets/tokens",
  "prompt injection via repo files",
  "auth bypass",
  "unauthorized API access",
  "billing logic bypass",
  "CI workflow tampering",
  "unsafe deserialization",
  "SQL/command injection"
]}, open(out,"w"), indent=2)
print(f"security-scan hard_fail={hard} hits={len(hits)}", file=sys.stderr)
PY