Exploitability Context for CVE Advisories (#89)

* feat(advisories): add exploitability context for CVE advisories * fix(ci): align exploitability workflow with signing model * docs(skills): add patch release changelog entries * chore(clawsec-feed): bump version to 0.0.5 * chore(clawsec-suite): bump version to 0.1.4 * fix(clawsec-nanoclaw): align exploitability handling and nanoclaw integration * chore(clawsec-nanoclaw): bump version to 0.0.2 * refactor(scripts): share feed path and mirror sync helpers * refactor(utils): unify cvss vector parsing flow * refactor(clawsec-nanoclaw): centralize advisory risk evaluation * docs(exploitability): refresh release metadata dates * fix(review): align feed signing and advisory dedupe * chore(clawsec-feed): bump version to 0.0.6 * chore(clawsec-nanoclaw): bump version to 0.0.3 * fix(backfill): limit signing to target feed only * fix(review): keep skill runtime verify-only and dedupe matching * chore(clawsec-nanoclaw): bump version to 0.0.4 * chore(skills): align versions with published tags * feat(feed): enrich local population with exploitability analysis * docs(exploitability): mark backfill as historical flow
2026-06-13 05:28:02 +03:00 · 2026-03-01 18:43:24 +02:00
parent 382db82483
commit 073e771b73
26 changed files with 2015 additions and 197 deletions
@@ -193,6 +193,74 @@ jobs:
          echo "Created advisory JSON:"
          cat tmp_advisory.json

+      - name: Set up Python for exploitability analysis
+        if: steps.parse.outputs.already_exists != 'true'
+        uses: actions/setup-python@8d2f52b6169cf4c0f64d2e9f6f8f6d6a6e1c90f7 # v5.4.1
+        with:
+          python-version: '3.10'
+
+      - name: Analyze exploitability for community advisory
+        if: steps.parse.outputs.already_exists != 'true'
+        run: |
+          set -euo pipefail
+
+          echo "=== Analyzing exploitability for community advisory ==="
+
+          # Extract fields from advisory for analysis
+          CVE_ID=$(jq -r '.id' tmp_advisory.json)
+          SEVERITY=$(jq -r '.severity // "medium"' tmp_advisory.json)
+          VULN_TYPE=$(jq -r '.type // "unknown"' tmp_advisory.json)
+          DESCRIPTION=$(jq -r '.description // ""' tmp_advisory.json)
+          REFERENCES=$(jq -c '.references // []' tmp_advisory.json)
+
+          # Map severity to approximate CVSS score for analysis
+          case "$SEVERITY" in
+            critical) CVSS_SCORE=9.5 ;;
+            high) CVSS_SCORE=7.5 ;;
+            medium) CVSS_SCORE=5.5 ;;
+            low) CVSS_SCORE=3.0 ;;
+            *) CVSS_SCORE=5.0 ;;
+          esac
+
+          # Build input JSON for analyzer
+          INPUT_JSON=$(jq -n \
+            --arg cve_id "$CVE_ID" \
+            --argjson cvss_score "$CVSS_SCORE" \
+            --arg type "$VULN_TYPE" \
+            --arg description "$DESCRIPTION" \
+            --argjson references "$REFERENCES" \
+            '{
+              cve_id: $cve_id,
+              cvss_score: $cvss_score,
+              type: $type,
+              description: $description,
+              references: $references
+            }')
+
+          # Run exploitability analysis with exploit detection.
+          # Continue without enrichment if analysis fails.
+          if ANALYSIS=$(echo "$INPUT_JSON" | python utils/analyze_exploitability.py --json --check-exploits 2>/dev/null); then
+            echo "$ANALYSIS" > tmp_exploitability.json
+            echo "✓ Analyzed $CVE_ID"
+
+            # Merge exploitability analysis into advisory
+            jq --slurpfile analysis tmp_exploitability.json '
+              . + {
+                exploitability_score: $analysis[0].exploitability_score,
+                exploitability_rationale: $analysis[0].exploitability_rationale,
+                attack_vector_analysis: $analysis[0].attack_vector_analysis,
+                exploit_detection: $analysis[0].exploit_detection
+              }
+            ' tmp_advisory.json > tmp_advisory_enriched.json
+
+            mv tmp_advisory_enriched.json tmp_advisory.json
+
+            echo "=== Exploitability analysis complete ==="
+            echo "Exploitability score: $(jq -r '.exploitability_score // "unknown"' tmp_advisory.json)"
+          else
+            echo "::warning::Failed to analyze exploitability for $CVE_ID, continuing without enrichment"
+          fi
+
      - name: Update feed
        if: steps.parse.outputs.already_exists != 'true'
        run: |
@@ -353,7 +353,9 @@ jobs:
              title: (.cve.descriptions[] | select(.lang == "en") | .value | .[0:100] + (if length > 100 then "..." else "" end)),
              affected: normalized_affected,
              platforms: normalized_platforms,
-              references: [.cve.references[]?.url // empty] | unique | .[0:3]
+              references: [.cve.references[]?.url // empty] | unique | .[0:3],
+              exploitability_score: null,
+              exploitability_rationale: null
            }]
          ' tmp/filtered_cves.json > tmp/nvd_current_state.json

@@ -568,7 +570,9 @@ jobs:
                published: .cve.published,
                references: [.cve.references[]?.url // empty] | unique | .[0:3],
                cvss_score: get_cvss_score,
-                nvd_url: ("https://nvd.nist.gov/vuln/detail/" + .cve.id)
+                nvd_url: ("https://nvd.nist.gov/vuln/detail/" + .cve.id),
+                exploitability_score: null,
+                exploitability_rationale: null
              }
            ]
          ' tmp/filtered_cves.json > tmp/new_advisories.json
@@ -582,6 +586,102 @@ jobs:
            jq '.[].id' tmp/new_advisories.json
          fi

+      - name: Set up Python for exploitability analysis
+        if: steps.transform.outputs.new_count != '0'
+        uses: actions/setup-python@8d2f52b6169cf4c0f64d2e9f6f8f6d6a6e1c90f7 # v5.4.1
+        with:
+          python-version: '3.10'
+
+      - name: Analyze exploitability for new advisories
+        if: steps.transform.outputs.new_count != '0'
+        run: |
+          set -euo pipefail
+
+          echo "=== Analyzing exploitability for new advisories ==="
+
+          # Extract CVSS vectors from filtered CVEs to merge with advisories
+          jq '
+            [.[] | {
+              id: .cve.id,
+              cvss_vector: (
+                .cve.metrics.cvssMetricV31[0]?.cvssData.vectorString //
+                .cve.metrics.cvssMetricV30[0]?.cvssData.vectorString //
+                .cve.metrics.cvssMetricV2[0]?.vectorString //
+                ""
+              )
+            }] | map({(.id): .cvss_vector}) | add
+          ' tmp/filtered_cves.json > tmp/cvss_vectors.json
+
+          # Process each advisory through exploitability analyzer
+          jq -c '.[]' tmp/new_advisories.json | while IFS= read -r advisory; do
+            CVE_ID=$(echo "$advisory" | jq -r '.id')
+            CVSS_SCORE=$(echo "$advisory" | jq -r '.cvss_score // 0')
+            CVSS_VECTOR=$(jq -r --arg id "$CVE_ID" '.[$id] // ""' tmp/cvss_vectors.json)
+            VULN_TYPE=$(echo "$advisory" | jq -r '.type // ""')
+            DESCRIPTION=$(echo "$advisory" | jq -r '.description // ""')
+            REFERENCES=$(echo "$advisory" | jq -c '.references // []')
+
+            # Build input JSON for analyzer
+            INPUT_JSON=$(jq -n \
+              --arg cve_id "$CVE_ID" \
+              --argjson cvss_score "$CVSS_SCORE" \
+              --arg cvss_vector "$CVSS_VECTOR" \
+              --arg type "$VULN_TYPE" \
+              --arg description "$DESCRIPTION" \
+              --argjson references "$REFERENCES" \
+              '{
+                cve_id: $cve_id,
+                cvss_score: $cvss_score,
+                cvss_vector: $cvss_vector,
+                type: $type,
+                description: $description,
+                references: $references
+              }')
+
+            # Run exploitability analysis with exploit detection.
+            # Keep processing if any single advisory analysis fails.
+            if ANALYSIS=$(echo "$INPUT_JSON" | python utils/analyze_exploitability.py --json --check-exploits 2>/dev/null); then
+              echo "$ANALYSIS" > "tmp/exploitability_${CVE_ID}.json"
+              echo "✓ Analyzed $CVE_ID"
+            else
+              echo "::warning::Failed to analyze exploitability for $CVE_ID, skipping enrichment"
+            fi
+          done
+
+          # Merge exploitability analysis back into advisories.
+          if ls tmp/exploitability_*.json >/dev/null 2>&1; then
+            jq -s '.' tmp/exploitability_*.json > tmp/exploitability_analyses.json
+          else
+            echo '[]' > tmp/exploitability_analyses.json
+          fi
+
+          jq --slurpfile analyses tmp/exploitability_analyses.json '
+            map(
+              . as $advisory |
+              ($analyses[0] | map(select(.cve_id == $advisory.id)) | first) as $analysis |
+              if $analysis then
+                $advisory + {
+                  exploitability_score: $analysis.exploitability_score,
+                  exploitability_rationale: $analysis.exploitability_rationale,
+                  attack_vector_analysis: $analysis.attack_vector_analysis,
+                  exploit_detection: $analysis.exploit_detection
+                }
+              else
+                $advisory
+              end
+            )
+          ' tmp/new_advisories.json > tmp/new_advisories_enriched.json
+
+          # Replace original with enriched version
+          mv tmp/new_advisories_enriched.json tmp/new_advisories.json
+
+          echo "=== Exploitability analysis complete ==="
+
+          # Show summary of exploitability scores
+          echo "Exploitability score distribution:"
+          jq -r '.[] | "\(.id): \(.exploitability_score // "unknown")"' tmp/new_advisories.json | \
+            awk -F': ' '{scores[$2]++} END {for (s in scores) print "  " s ": " scores[s]}'
+
      - name: Update feed.json
        if: steps.transform.outputs.new_count != '0' || steps.updates.outputs.update_count != '0'
        run: |
@@ -203,6 +203,17 @@ The feed polls CVEs related to:
 - Prompt injection patterns
 - Agent security vulnerabilities

+### Exploitability Context
+
+ClawSec enriches CVE advisories with **exploitability context** to help agents assess real-world risk beyond raw CVSS scores. Newly analyzed advisories can include:
+
+- **Exploit Evidence**: Whether public exploits exist in the wild
+- **Weaponization Status**: If exploits are integrated into common attack frameworks
+- **Attack Requirements**: Prerequisites needed for successful exploitation (network access, authentication, user interaction)
+- **Risk Assessment**: Contextualized risk level combining technical severity with exploitability
+
+This feature helps agents prioritize vulnerabilities that pose immediate threats versus theoretical risks, enabling smarter security decisions.
+
 ### Advisory Schema

 **NVD CVE Advisory:**
@@ -217,6 +228,8 @@ The feed polls CVEs related to:
  "published": "2026-02-01T00:00:00Z",
  "cvss_score": 8.8,
  "nvd_url": "https://nvd.nist.gov/vuln/detail/CVE-2026-XXXXX",
+  "exploitability_score": "high|medium|low|unknown",
+  "exploitability_rationale": "Why this CVE is or is not likely exploitable in agent deployments",
  "references": ["..."],
  "action": "Recommended remediation"
 }
@@ -0,0 +1,281 @@
+#!/bin/bash
+# backfill-exploitability.sh
+# Adds exploitability scoring to existing advisories in feed.json that don't have it yet.
+# Historical maintenance utility: normal advisory generation should use
+# poll-nvd workflow (init/reset when rebuilding) or populate-local-feed.sh.
+#
+# Usage: ./scripts/backfill-exploitability.sh [--dry-run] [--feed PATH]
+#   --dry-run   Show what would be updated without making changes
+#   --feed PATH Use specified feed file (default: advisories/feed.json)
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+# shellcheck source=./feed-utils.sh
+source "$SCRIPT_DIR/feed-utils.sh"
+
+# Configuration
+init_feed_paths "$PROJECT_ROOT"
+ANALYZER="$PROJECT_ROOT/utils/analyze_exploitability.py"
+SIGNING_PRIVATE_KEY="${CLAWSEC_FEED_SIGNING_PRIVATE_KEY_PATH:-${CLAWSEC_SIGNING_PRIVATE_KEY_PATH:-$PROJECT_ROOT/clawsec-signing-private.pem}}"
+SIGNING_PUBLIC_KEY="${CLAWSEC_FEED_SIGNING_PUBLIC_KEY_PATH:-${CLAWSEC_SIGNING_PUBLIC_KEY_PATH:-$PROJECT_ROOT/clawsec-signing-public.pem}}"
+SIGNING_PASSPHRASE="${CLAWSEC_FEED_SIGNING_PRIVATE_KEY_PASSPHRASE:-${CLAWSEC_SIGNING_PRIVATE_KEY_PASSPHRASE:-}}"
+
+sign_and_verify_feed_signature() {
+  local feed_file="$1"
+  local signature_file="$2"
+  local tmp_dir
+  local tmp_signature
+  local signature_bin
+  local passin_file
+
+  tmp_dir=$(mktemp -d)
+  tmp_signature="${signature_file}.tmp.$$"
+  signature_bin="$tmp_dir/signature.bin"
+  passin_file="$tmp_dir/passin.txt"
+
+  if [ -n "$SIGNING_PASSPHRASE" ]; then
+    printf '%s' "$SIGNING_PASSPHRASE" > "$passin_file"
+    chmod 600 "$passin_file"
+
+    if ! openssl pkeyutl -sign -rawin -inkey "$SIGNING_PRIVATE_KEY" -passin "file:$passin_file" -in "$feed_file" \
+      | openssl base64 -A > "$tmp_signature"; then
+      rm -rf "$tmp_dir"
+      rm -f "$tmp_signature"
+      echo "Error: Failed to sign $feed_file" >&2
+      return 1
+    fi
+  elif ! openssl pkeyutl -sign -rawin -inkey "$SIGNING_PRIVATE_KEY" -in "$feed_file" \
+    | openssl base64 -A > "$tmp_signature"; then
+    rm -rf "$tmp_dir"
+    rm -f "$tmp_signature"
+    echo "Error: Failed to sign $feed_file" >&2
+    return 1
+  fi
+
+  if ! openssl base64 -d -A -in "$tmp_signature" -out "$signature_bin"; then
+    rm -rf "$tmp_dir"
+    rm -f "$tmp_signature"
+    echo "Error: Failed to decode generated signature for $feed_file" >&2
+    return 1
+  fi
+
+  if ! openssl pkeyutl -verify -rawin -pubin -inkey "$SIGNING_PUBLIC_KEY" -sigfile "$signature_bin" -in "$feed_file" >/dev/null; then
+    rm -rf "$tmp_dir"
+    rm -f "$tmp_signature"
+    echo "Error: Signature verification failed after signing $feed_file" >&2
+    return 1
+  fi
+
+  mv "$tmp_signature" "$signature_file"
+  rm -rf "$tmp_dir"
+  echo "✓ Re-signed and verified: $signature_file"
+}
+
+# Parse args
+DRY_RUN=false
+REQUIRE_SIGNING=false
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+    --dry-run)
+      DRY_RUN=true
+      shift
+      ;;
+    --feed)
+      FEED_PATH="$2"
+      shift 2
+      ;;
+    *)
+      echo "Unknown option: $1"
+      echo "Usage: $0 [--dry-run] [--feed PATH]"
+      exit 1
+      ;;
+  esac
+done
+
+echo "=== ClawSec Exploitability Backfill ==="
+echo "Feed path: $FEED_PATH"
+echo "Dry run: $DRY_RUN"
+echo ""
+
+# Verify prerequisites
+if [ ! -f "$FEED_PATH" ]; then
+  echo "Error: Feed file not found: $FEED_PATH"
+  exit 1
+fi
+
+if [ ! -f "$ANALYZER" ]; then
+  echo "Error: Analyzer script not found: $ANALYZER"
+  exit 1
+fi
+
+# Check Python availability
+if ! command -v python3 &> /dev/null; then
+  echo "Error: python3 is required but not found in PATH"
+  exit 1
+fi
+
+# Verify analyzer works
+if ! python3 "$ANALYZER" --help &> /dev/null; then
+  echo "Error: Analyzer script failed to run. Check Python environment."
+  exit 1
+fi
+
+# Determine whether detached signatures must be regenerated.
+# Runtime agents that only have public keys should run in dry-run mode.
+if [ "$DRY_RUN" = "false" ]; then
+  if [ -f "${FEED_PATH}.sig" ]; then
+    REQUIRE_SIGNING=true
+  fi
+fi
+
+if [ "$REQUIRE_SIGNING" = "true" ]; then
+  if ! command -v openssl &> /dev/null; then
+    echo "Error: openssl is required for detached signature signing/verification"
+    exit 1
+  fi
+
+  if [ ! -f "$SIGNING_PRIVATE_KEY" ]; then
+    echo "Error: Signing private key not found: $SIGNING_PRIVATE_KEY"
+    echo "This backfill updates signed feed artifacts. Use --dry-run in public-key-only environments."
+    exit 1
+  fi
+
+  if [ ! -f "$SIGNING_PUBLIC_KEY" ]; then
+    echo "Error: Signing public key not found: $SIGNING_PUBLIC_KEY"
+    exit 1
+  fi
+fi
+
+# Create temp directory
+TEMP_DIR=$(mktemp -d)
+trap 'rm -rf "$TEMP_DIR"' EXIT
+
+echo "=== Analyzing Feed ==="
+
+# Extract advisories without exploitability_score
+jq '.advisories | map(select(.exploitability_score == null or .exploitability_score == ""))' \
+  "$FEED_PATH" > "$TEMP_DIR/missing_exploitability.json"
+
+MISSING_COUNT=$(jq 'length' "$TEMP_DIR/missing_exploitability.json")
+TOTAL_COUNT=$(jq '.advisories | length' "$FEED_PATH")
+ALREADY_DONE=$((TOTAL_COUNT - MISSING_COUNT))
+
+echo "Total advisories: $TOTAL_COUNT"
+echo "Already have exploitability: $ALREADY_DONE"
+echo "Missing exploitability: $MISSING_COUNT"
+echo ""
+
+if [ "$MISSING_COUNT" -eq 0 ]; then
+  echo "✓ All advisories already have exploitability scores!"
+  exit 0
+fi
+
+if [ "$DRY_RUN" = "true" ]; then
+  echo "=== Dry Run - Would Update These Advisories ==="
+  jq -r '.[] | .id' "$TEMP_DIR/missing_exploitability.json"
+  echo ""
+  echo "Total advisories to update: $MISSING_COUNT"
+  exit 0
+fi
+
+echo "=== Processing Advisories ==="
+
+# Process each advisory
+PROCESSED=0
+FAILED=0
+
+# Read original feed to preserve all metadata
+cp "$FEED_PATH" "$TEMP_DIR/feed_working.json"
+
+while IFS= read -r advisory; do
+  CVE_ID=$(echo "$advisory" | jq -r '.id')
+
+  echo -n "Processing $CVE_ID... "
+
+  # Prepare input for analyzer
+  ANALYZER_INPUT=$(echo "$advisory" | jq '{
+    cve_id: .id,
+    cvss_score: (.cvss_score // 0.0),
+    type: .type,
+    description: .description,
+    references: (.references // [])
+  }')
+
+  # Run analyzer
+  if ANALYSIS=$(echo "$ANALYZER_INPUT" | python3 "$ANALYZER" --json --check-exploits 2>/dev/null); then
+    # Extract exploitability fields
+    EXPL_SCORE=$(echo "$ANALYSIS" | jq -r '.exploitability_score // "unknown"')
+    EXPL_RATIONALE=$(echo "$ANALYSIS" | jq -r '.exploitability_rationale // "No rationale available"')
+
+    # Update advisory in working feed
+    jq --arg id "$CVE_ID" \
+       --arg score "$EXPL_SCORE" \
+       --arg rationale "$EXPL_RATIONALE" \
+       '(.advisories[] | select(.id == $id)) |= (. + {
+         exploitability_score: $score,
+         exploitability_rationale: $rationale
+       })' "$TEMP_DIR/feed_working.json" > "$TEMP_DIR/feed_updated.json"
+
+    mv "$TEMP_DIR/feed_updated.json" "$TEMP_DIR/feed_working.json"
+
+    echo "✓ $EXPL_SCORE"
+    PROCESSED=$((PROCESSED + 1))
+  else
+    echo "✗ Failed"
+    FAILED=$((FAILED + 1))
+  fi
+done < <(jq -c '.[]' "$TEMP_DIR/missing_exploitability.json")
+
+# Check if loop executed successfully
+if [ ! -f "$TEMP_DIR/feed_working.json" ]; then
+  echo "Error: Feed processing failed"
+  exit 1
+fi
+
+echo ""
+echo "=== Processing Complete ==="
+echo "Processed: $PROCESSED"
+echo "Failed: $FAILED"
+echo ""
+
+# Write updated feed
+echo "Writing updated feed to: $FEED_PATH"
+cp "$TEMP_DIR/feed_working.json" "$FEED_PATH"
+
+# Update feed version and timestamp
+CURRENT_VERSION=$(jq -r '.version' "$FEED_PATH")
+UPDATED_TS=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+
+jq --arg ts "$UPDATED_TS" '.updated = $ts' "$FEED_PATH" > "$TEMP_DIR/feed_final.json"
+mv "$TEMP_DIR/feed_final.json" "$FEED_PATH"
+
+echo "✓ Updated feed version: $CURRENT_VERSION"
+echo "✓ Updated timestamp: $UPDATED_TS"
+echo ""
+
+if [ "$REQUIRE_SIGNING" = "true" ]; then
+  echo ""
+  echo "=== Re-signing Advisory Feed ==="
+
+  if [ -f "${FEED_PATH}.sig" ]; then
+    if ! sign_and_verify_feed_signature "$FEED_PATH" "${FEED_PATH}.sig"; then
+      exit 1
+    fi
+  fi
+fi
+
+echo ""
+echo "=== Summary ==="
+echo "✓ Backfill complete!"
+echo "✓ $PROCESSED advisories updated with exploitability scores"
+
+if [ "$FAILED" -gt 0 ]; then
+  echo "⚠ $FAILED advisories failed analysis (kept original data)"
+fi
+
+# Verify final state
+FINAL_MISSING=$(jq '.advisories | map(select(.exploitability_score == null or .exploitability_score == "")) | length' "$FEED_PATH")
+echo "✓ Advisories still missing exploitability: $FINAL_MISSING"
@@ -0,0 +1,37 @@
+#!/bin/bash
+# feed-utils.sh
+# Shared advisory feed path and sync helpers for local/maintenance scripts.
+
+init_feed_paths() {
+  local project_root="$1"
+
+  : "${FEED_PATH:=$project_root/advisories/feed.json}"
+  : "${SKILL_FEED_PATH:=$project_root/skills/clawsec-feed/advisories/feed.json}"
+  : "${PUBLIC_FEED_PATH:=$project_root/public/advisories/feed.json}"
+}
+
+sync_feed_to_mirrors() {
+  local source_feed="$1"
+  local mode="${2:-create}"
+
+  local target
+  for target in "$SKILL_FEED_PATH" "$PUBLIC_FEED_PATH"; do
+    case "$mode" in
+      create)
+        mkdir -p "$(dirname "$target")"
+        cp "$source_feed" "$target"
+        echo "✓ Updated: $target"
+        ;;
+      existing-only)
+        if [ -f "$target" ]; then
+          cp "$source_feed" "$target"
+          echo "✓ Updated: $target"
+        fi
+        ;;
+      *)
+        echo "Error: unsupported mirror sync mode: $mode" >&2
+        return 1
+        ;;
+    esac
+  done
+}
@@ -11,13 +11,14 @@ set -euo pipefail

 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+# shellcheck source=./feed-utils.sh
+source "$SCRIPT_DIR/feed-utils.sh"

 # Configuration - same as pipeline
-FEED_PATH="$PROJECT_ROOT/advisories/feed.json"
-SKILL_FEED_PATH="$PROJECT_ROOT/skills/clawsec-feed/advisories/feed.json"
-PUBLIC_FEED_PATH="$PROJECT_ROOT/public/advisories/feed.json"
+init_feed_paths "$PROJECT_ROOT"
 KEYWORDS="OpenClaw clawdbot Moltbot NanoClaw WhatsApp-bot baileys"
 GITHUB_REF_PATTERN="github.com/openclaw/openclaw github.com/qwibitai/NanoClaw"
+ANALYZER="$PROJECT_ROOT/utils/analyze_exploitability.py"

 # Parse args
 DAYS_BACK=120
@@ -46,6 +47,22 @@ echo "Days back: $DAYS_BACK"
 echo "Force mode: $FORCE"
 echo ""

+# Verify exploitability analyzer prerequisites
+if ! command -v python3 &> /dev/null; then
+  echo "Error: python3 is required but not found in PATH"
+  exit 1
+fi
+
+if [ ! -f "$ANALYZER" ]; then
+  echo "Error: Exploitability analyzer not found: $ANALYZER"
+  exit 1
+fi
+
+if ! python3 "$ANALYZER" --help &> /dev/null; then
+  echo "Error: Exploitability analyzer failed to run. Check Python environment."
+  exit 1
+fi
+
 # Create temp directory
 TEMP_DIR=$(mktemp -d)
 trap 'rm -rf "$TEMP_DIR"' EXIT
@@ -62,7 +79,7 @@ fi
 if [ -z "${START_DATE:-}" ]; then
  # macOS vs Linux date compatibility
  if date -v-1d > /dev/null 2>&1; then
-    START_DATE=$(date -u -v-${DAYS_BACK}d +%Y-%m-%dT%H:%M:%S.000Z)
+    START_DATE=$(date -u -v-"${DAYS_BACK}"d +%Y-%m-%dT%H:%M:%S.000Z)
  else
    START_DATE=$(date -u -d "${DAYS_BACK} days ago" +%Y-%m-%dT%H:%M:%S.000Z)
  fi
@@ -74,8 +91,8 @@ echo "End date: $END_DATE"
 echo ""

 # URL encode dates
-START_ENC=$(echo "$START_DATE" | sed 's/:/%3A/g')
-END_ENC=$(echo "$END_DATE" | sed 's/:/%3A/g')
+START_ENC=${START_DATE//:/%3A}
+END_ENC=${END_DATE//:/%3A}

 echo "=== Fetching CVEs from NVD ==="

@@ -281,7 +298,9 @@ jq --argjson existing "$EXISTING_JSON" '
      published: .cve.published,
      references: [.cve.references[]?.url // empty] | unique | .[0:3],
      cvss_score: get_cvss_score,
-      nvd_url: ("https://nvd.nist.gov/vuln/detail/" + .cve.id)
+      nvd_url: ("https://nvd.nist.gov/vuln/detail/" + .cve.id),
+      exploitability_score: null,
+      exploitability_rationale: null
    }
  ]
 ' "$TEMP_DIR/filtered_cves.json" > "$TEMP_DIR/new_advisories.json"
@@ -296,6 +315,87 @@ if [ "$NEW_COUNT" -eq 0 ]; then
  exit 0
 fi

+echo ""
+echo "=== Analyzing Exploitability ==="
+
+# Build CVSS vector lookup for enriched analysis inputs.
+jq '
+  [.[] | {
+    id: .cve.id,
+    cvss_vector: (
+      .cve.metrics.cvssMetricV31[0]?.cvssData.vectorString //
+      .cve.metrics.cvssMetricV30[0]?.cvssData.vectorString //
+      .cve.metrics.cvssMetricV2[0]?.vectorString //
+      ""
+    )
+  }] | map({(.id): .cvss_vector}) | add
+' "$TEMP_DIR/filtered_cves.json" > "$TEMP_DIR/cvss_vectors.json"
+
+ANALYZED_COUNT=0
+FAILED_ANALYSIS=0
+
+while IFS= read -r advisory; do
+  CVE_ID=$(echo "$advisory" | jq -r '.id')
+  CVSS_SCORE=$(echo "$advisory" | jq -r '.cvss_score // 0')
+  CVSS_VECTOR=$(jq -r --arg id "$CVE_ID" '.[$id] // ""' "$TEMP_DIR/cvss_vectors.json")
+  VULN_TYPE=$(echo "$advisory" | jq -r '.type // ""')
+  DESCRIPTION=$(echo "$advisory" | jq -r '.description // ""')
+  REFERENCES=$(echo "$advisory" | jq -c '.references // []')
+
+  INPUT_JSON=$(jq -n \
+    --arg cve_id "$CVE_ID" \
+    --argjson cvss_score "$CVSS_SCORE" \
+    --arg cvss_vector "$CVSS_VECTOR" \
+    --arg type "$VULN_TYPE" \
+    --arg description "$DESCRIPTION" \
+    --argjson references "$REFERENCES" \
+    '{
+      cve_id: $cve_id,
+      cvss_score: $cvss_score,
+      cvss_vector: $cvss_vector,
+      type: $type,
+      description: $description,
+      references: $references
+    }')
+
+  if ANALYSIS=$(echo "$INPUT_JSON" | python3 "$ANALYZER" --json --check-exploits 2>/dev/null); then
+    echo "$ANALYSIS" > "$TEMP_DIR/exploitability_${CVE_ID}.json"
+    SCORE=$(echo "$ANALYSIS" | jq -r '.exploitability_score // "unknown"')
+    echo "  ✓ $CVE_ID -> $SCORE"
+    ANALYZED_COUNT=$((ANALYZED_COUNT + 1))
+  else
+    echo "  ⚠ $CVE_ID analysis failed; keeping null exploitability fields"
+    FAILED_ANALYSIS=$((FAILED_ANALYSIS + 1))
+  fi
+done < <(jq -c '.[]' "$TEMP_DIR/new_advisories.json")
+
+if ls "$TEMP_DIR"/exploitability_*.json >/dev/null 2>&1; then
+  jq -s '.' "$TEMP_DIR"/exploitability_*.json > "$TEMP_DIR/exploitability_analyses.json"
+else
+  echo '[]' > "$TEMP_DIR/exploitability_analyses.json"
+fi
+
+jq --slurpfile analyses "$TEMP_DIR/exploitability_analyses.json" '
+  map(
+    . as $advisory |
+    ($analyses[0] | map(select(.cve_id == $advisory.id)) | first) as $analysis |
+    if $analysis then
+      $advisory + {
+        exploitability_score: $analysis.exploitability_score,
+        exploitability_rationale: $analysis.exploitability_rationale,
+        attack_vector_analysis: $analysis.attack_vector_analysis,
+        exploit_detection: $analysis.exploit_detection
+      }
+    else
+      $advisory
+    end
+  )
+' "$TEMP_DIR/new_advisories.json" > "$TEMP_DIR/new_advisories_enriched.json"
+
+mv "$TEMP_DIR/new_advisories_enriched.json" "$TEMP_DIR/new_advisories.json"
+
+echo "Exploitability analysis complete: $ANALYZED_COUNT analyzed, $FAILED_ANALYSIS failed"
+
 echo ""
 echo "=== New Advisories ==="
 jq -r '.[] | "  \(.id) [\(.severity)] - \(.title)"' "$TEMP_DIR/new_advisories.json"
@@ -339,15 +439,8 @@ if jq empty "$TEMP_DIR/updated_feed.json" 2>/dev/null; then
  cp "$TEMP_DIR/updated_feed.json" "$FEED_PATH"
  echo "✓ Updated: $FEED_PATH"

-  # Update skill feed
-  mkdir -p "$(dirname "$SKILL_FEED_PATH")"
-  cp "$FEED_PATH" "$SKILL_FEED_PATH"
-  echo "✓ Updated: $SKILL_FEED_PATH"
-  
-  # Update public feed for local dev
-  mkdir -p "$(dirname "$PUBLIC_FEED_PATH")"
-  cp "$FEED_PATH" "$PUBLIC_FEED_PATH"
-  echo "✓ Updated: $PUBLIC_FEED_PATH"
+  # Sync feed mirrors for local skill/public consumers.
+  sync_feed_to_mirrors "$FEED_PATH" "create"
  
  echo ""
  TOTAL_ADVISORIES=$(jq '.advisories | length' "$FEED_PATH")
@@ -76,13 +76,13 @@ fi
 # ESLint
 echo -e "\n${YELLOW}Running ESLint...${NC}"
 if $FIX_MODE; then
-  if npx eslint . --ext .ts,.tsx,.js,.jsx,.mjs --fix; then
+  if npx eslint . --ext .ts,.tsx,.js,.jsx,.mjs --ignore-pattern '.auto-claude/**' --fix; then
    check_pass "ESLint (with auto-fix)"
  else
    check_fail "ESLint found unfixable issues"
  fi
 else
-  if npx eslint . --ext .ts,.tsx,.js,.jsx,.mjs --max-warnings 0; then
+  if npx eslint . --ext .ts,.tsx,.js,.jsx,.mjs --ignore-pattern '.auto-claude/**' --max-warnings 0; then
    check_pass "ESLint"
  else
    check_fail "ESLint found issues (run with --fix to auto-fix)"
@@ -190,7 +190,7 @@ print_header "Security"
 # Trivy FS Scan
 if command -v trivy &> /dev/null; then
  echo -e "\n${YELLOW}Running Trivy filesystem scan...${NC}"
-  if trivy fs . --severity CRITICAL,HIGH --exit-code 1 --ignore-unfixed; then
+  if trivy fs . --severity CRITICAL,HIGH --exit-code 1 --ignore-unfixed --skip-dirs .auto-claude --skip-files clawsec-signing-private.pem; then
    check_pass "Trivy filesystem scan"
  else
    check_fail "Trivy found CRITICAL/HIGH vulnerabilities"
@@ -0,0 +1,19 @@
+# Changelog
+
+All notable changes to the ClawSec Feed skill will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.0.5] - 2026-02-28
+
+### Added
+
+- Exploitability-focused advisory guidance, including filtering and prioritization examples.
+- Notification examples that include exploitability context and rationale.
+
+### Changed
+
+- Clarified exploitability scoring guidance to match runtime values (`high|medium|low|unknown`).
+- Updated response-priority guidance to align with exploitability-first triage.
+- De-duplicated exploitability filtering guidance in `SKILL.md` by pointing to canonical docs in `wiki/exploitability-scoring.md` and `clawsec-suite`.
@@ -1,6 +1,6 @@
 ---
 name: clawsec-feed
-version: 0.0.4
+version: 0.0.5
 description: Security advisory feed with automated NVD CVE polling for OpenClaw-related vulnerabilities. Updated daily.
 homepage: https://clawsec.prompt.security
 metadata: {"openclaw":{"emoji":"📡","category":"security"}}
@@ -318,7 +318,9 @@ curl -sSL --fail --show-error --retry 3 --retry-delay 1 "$FEED_URL"
      "description": "Skill sends user data to external server",
      "affected": ["helper-plus@1.0.0", "helper-plus@1.0.1"],
      "action": "Remove immediately",
-      "published": "2026-02-01T10:00:00Z"
+      "published": "2026-02-01T10:00:00Z",
+      "exploitability_score": "critical",
+      "exploitability_rationale": "Trivially exploitable through normal skill usage; no special conditions required. Active exploitation observed in the wild."
    }
  ]
 }
@@ -385,6 +387,42 @@ fi
 echo "$RECENT"
 ```

+### Filter by exploitability score
+
+Shared exploitability prioritization guidance is maintained in:
+
+- [`wiki/exploitability-scoring.md`](../../wiki/exploitability-scoring.md)
+- [`skills/clawsec-suite/SKILL.md`](../clawsec-suite/SKILL.md) ("Quick feed check")
+
+### Get exploitability context for an advisory
+
+```bash
+# Show exploitability details for a specific CVE
+CVE_ID="CVE-2026-27488"
+echo "$FEED" | jq --arg cve "$CVE_ID" '.advisories[] | select(.id == $cve) | {
+  id: .id,
+  severity: .severity,
+  exploitability_score: .exploitability_score,
+  exploitability_rationale: .exploitability_rationale,
+  title: .title
+}'
+```
+
+### Prioritize advisories by exploitability
+
+```bash
+# Sort advisories by exploitability (critical → high → medium → low)
+# This helps agents focus on the most immediately actionable threats
+echo "$FEED" | jq '[.advisories[] | select(.exploitability_score != null)] |
+  sort_by(
+    if .exploitability_score == "critical" then 0
+    elif .exploitability_score == "high" then 1
+    elif .exploitability_score == "medium" then 2
+    elif .exploitability_score == "low" then 3
+    else 4 end
+  )'
+```
+
 ---

 ## Cross-Reference Installed Skills
@@ -476,23 +514,75 @@ done

 ---

+## Prioritizing High-Exploitability Threats
+
+**IMPORTANT:** When reviewing advisories, always prioritize by **exploitability score** in addition to severity. The exploitability score indicates how easily a vulnerability can be exploited in practice, helping you focus on the most actionable threats.
+
+### Exploitability Priority Levels
+
+| Exploitability | Meaning | Action Priority |
+|----------------|---------|-----------------|
+| `high` | Trivially or easily exploitable with public tooling | **Immediate notification** |
+| `medium` | Exploitable but requires specific conditions | **Standard notification** |
+| `low` | Difficult to exploit or theoretical | **Low priority notification** |
+
+### How to Use Exploitability in Notifications
+
+1. **Filter for high-exploitability first:**
+   ```bash
+   # Get high exploitability advisories
+   echo "$FEED" | jq '.advisories[] | select(.exploitability_score == "high")'
+   ```
+
+2. **Include exploitability in notifications:**
+   ```
+   📡 ClawSec Feed: High-exploitability alert
+
+   CRITICAL - CVE-2026-27488 (Exploitability: HIGH)
+     → Trivially exploitable RCE in skill-loader v2.1.0
+     → Public exploit code available
+     → Recommended action: Immediate removal or upgrade to v2.1.1
+   ```
+
+3. **Prioritize by both severity AND exploitability:**
+   - A HIGH severity + HIGH exploitability CVE is more urgent than a CRITICAL severity + LOW exploitability CVE
+   - Focus user attention on threats that are both severe and easily exploitable
+   - Include the exploitability rationale to help users understand the risk context
+
+### Example Notification Priority Order
+
+When multiple advisories exist, present them in this order:
+1. **Critical severity + High exploitability** - most urgent
+2. **High severity + High exploitability**
+3. **Critical severity + Medium/Low exploitability**
+4. **High severity + Medium/Low exploitability**
+5. **Medium/Low severity** (any exploitability)
+
+This ensures you alert users to the most actionable, immediately dangerous threats first.
+
+---
+
 ## When to Notify Your User

 **Notify Immediately (Critical):**
 - New critical advisory affecting an installed skill
 - Active exploitation detected
+- **High exploitability score** (regardless of severity)

 **Notify Soon (High):**
 - New high-severity advisory affecting installed skills
 - Failed to fetch advisory feed (network issue?)
+- Medium exploitability with high severity

 **Notify at Next Interaction (Medium):**
 - New medium-severity advisories
 - General security updates
+- Low exploitability advisories

 **Log Only (Low/Info):**
 - Low-severity advisories (mention if user asks)
 - Feed checked, no new advisories
+- Theoretical vulnerabilities (low exploitability, low severity)

 ---

@@ -503,11 +593,13 @@ done
 ```
 📡 ClawSec Feed: 2 new advisories since last check

-CRITICAL - GA-2026-015: Malicious prompt pattern "ignore-all"
+CRITICAL - GA-2026-015: Malicious prompt pattern "ignore-all" (Exploitability: HIGH)
  → Detected prompt injection technique. Update your system prompt defenses.
+  → Exploitability: Easily exploitable with publicly documented techniques.

-HIGH - GA-2026-016: Vulnerable skill "data-helper" v1.2.0
+HIGH - GA-2026-016: Vulnerable skill "data-helper" v1.2.0 (Exploitability: MEDIUM)
  → You have this installed! Recommended action: Update to v1.2.1 or remove.
+  → Exploitability: Requires specific configuration; not trivially exploitable.
 ```

 ### If nothing new:
@@ -1,6 +1,6 @@
 {
  "name": "clawsec-feed",
-  "version": "0.0.4",
+  "version": "0.0.5",
  "description": "Security advisory feed monitoring for AI agents. Subscribe to community-driven threat intelligence.",
  "author": "prompt-security",
  "license": "AGPL-3.0-or-later",
@@ -21,6 +21,11 @@
        "required": true,
        "description": "Advisory feed skill documentation"
      },
+      {
+        "path": "CHANGELOG.md",
+        "required": true,
+        "description": "Version history for advisory feed updates"
+      },
      {
        "path": "advisories/feed.json",
        "required": true,
@@ -0,0 +1,20 @@
+# Changelog
+
+All notable changes to the ClawSec NanoClaw compatibility skill will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.0.2] - 2026-02-28
+
+### Added
+
+- Exploitability-aware advisory output in NanoClaw MCP tools (`exploitability_score`, `exploitability_rationale`).
+- Exploitability filtering (`exploitabilityScore`) for `clawsec_list_advisories`.
+
+### Changed
+
+- Updated NanoClaw advisory sorting and pre-install safety recommendation logic to prioritize exploitability context.
+- Updated NanoClaw integration docs to match current host/container integration points (`src/ipc.ts`, `src/index.ts`) and current cache schema.
+- Removed duplicate exploitability normalization logic from MCP advisory tools and now reuse `normalizeExploitabilityScore` from `lib/risk.ts`.
+- Reused `matchesAffectedSpecifier` from `lib/advisories.ts` in MCP advisory tools to keep skill/version matching logic centralized and consistent.
@@ -8,7 +8,7 @@ ClawSec provides security advisory monitoring for NanoClaw through:
 - **MCP Tools**: Agents can check for vulnerabilities via `clawsec_check_advisories`
 - **Advisory Feed**: Automatic monitoring of https://clawsec.prompt.security/advisories/feed.json
 - **Signature Verification**: Ed25519-signed feeds ensure integrity
- **Platform Targeting**: Advisories can be NanoClaw-specific or cross-platform
+- **Exploitability Context**: Advisories include exploitability score and rationale for triage

 ## Prerequisites

@@ -57,18 +57,30 @@ in each tool file).

 Add the host-side IPC handlers for ClawSec operations.

-**File**: `host/ipc-handler.ts`
+**File**: `src/ipc.ts`

 ```typescript
-// Add this import at the top
-import { registerClawSecHandlers } from '../skills/clawsec-nanoclaw/host-services/ipc-handlers.js';
+// Add these imports at the top
+import { handleAdvisoryIpc } from '../skills/clawsec-nanoclaw/host-services/ipc-handlers.js';
+import { AdvisoryCacheManager } from '../skills/clawsec-nanoclaw/host-services/advisory-cache.js';
+import { SkillSignatureVerifier } from '../skills/clawsec-nanoclaw/host-services/skill-signature-handler.js';

-// In your IPC handler setup function
-export function setupIpcHandlers() {
-  // ... your existing handlers ...
+// Initialize these once in host startup and pass through deps
+const advisoryCacheManager = new AdvisoryCacheManager('/workspace/project/data', logger);
+const signatureVerifier = new SkillSignatureVerifier();

-  // Register ClawSec handlers
-  registerClawSecHandlers();
+// In processTaskIpc switch:
+case 'refresh_advisory_cache':
+case 'verify_skill_signature':
+  await handleAdvisoryIpc(
+    data,
+    { advisoryCacheManager, signatureVerifier },
+    logger,
+    sourceGroup
+  );
+  break;
+default:
+  // existing task handling
 }
 ```

@@ -76,23 +88,25 @@ export function setupIpcHandlers() {

 Add the advisory cache manager to your host services.

-**File**: `host/index.ts` (or your main entry point)
+**File**: `src/index.ts` (or your main entry point)

 ```typescript
-// Add this import
-import { startAdvisoryCache } from '../skills/clawsec-nanoclaw/host-services/advisory-cache.js';
+import { AdvisoryCacheManager } from '../skills/clawsec-nanoclaw/host-services/advisory-cache.js';

 // Start the service when your host process starts
 async function main() {
  // ... your existing initialization ...

-  // Start ClawSec advisory cache (fetches feed every 6 hours)
-  startAdvisoryCache({
-    cacheFile: '/workspace/project/data/clawsec-advisory-cache.json',
-    feedUrl: 'https://clawsec.prompt.security/advisories/feed.json',
-    publicKeyPath: '/workspace/project/skills/clawsec-nanoclaw/advisories/feed-signing-public.pem',
-    refreshInterval: 6 * 60 * 60 * 1000, // 6 hours
+  // Initialize cache manager and prime it at startup
+  const advisoryCacheManager = new AdvisoryCacheManager('/workspace/project/data', logger);
+  await advisoryCacheManager.initialize();
+
+  // Recommended refresh cadence (6h)
+  setInterval(() => {
+    advisoryCacheManager.refresh().catch((error) => {
+      logger.error({ error }, 'Periodic advisory cache refresh failed');
    });
+  }, 6 * 60 * 60 * 1000);

  // ... rest of your startup ...
 }
@@ -151,9 +165,9 @@ cat /workspace/project/data/clawsec-advisory-cache.json

 You should see:
 - `feed`: Array of advisories
- `signature`: Ed25519 signature
- `lastFetch`: Timestamp of last update
+- `fetchedAt`: Timestamp of last update
 - `verified`: Should be `true`
+- `publicKeyFingerprint`: SHA-256 fingerprint of the pinned signing key

 ## Usage Examples

@@ -183,13 +197,13 @@ You can also call the MCP tools directly from agent code:
 ```typescript
 // Check all installed skills
 const result = await tools.clawsec_check_advisories({
-  skillsRoot: '/workspace/project/skills'
+  installRoot: '/home/node/.claude/skills'
 });

 // Check specific skill before installation
 const safetyCheck = await tools.clawsec_check_skill_safety({
  skillName: 'risky-skill',
-  version: '1.0.0'
+  skillVersion: '1.0.0'
 });
 ```

@@ -199,19 +213,19 @@ const safetyCheck = await tools.clawsec_check_skill_safety({

 Default: `/workspace/project/data/clawsec-advisory-cache.json`

-To change, update the `cacheFile` parameter in `startAdvisoryCache()`.
+To change, pass a different data directory path to `new AdvisoryCacheManager(dataDir, logger)`.

 ### Refresh Interval

 Default: 6 hours

-To change, update the `refreshInterval` parameter (in milliseconds).
+To change, update the `setInterval(...)` duration (in milliseconds) in host startup.

 ### Feed URL

 Default: `https://clawsec.prompt.security/advisories/feed.json`

-To use a mirror or custom feed, update the `feedUrl` parameter.
+To use a mirror or custom feed, update `FEED_URL` in `skills/clawsec-nanoclaw/host-services/advisory-cache.ts`.

 ## Platform-Specific Advisories

@@ -222,7 +236,7 @@ ClawSec advisories can target specific platforms:
 - **`platforms: ["openclaw", "nanoclaw"]`**: Affects both
 - **No `platforms` field**: Applies to all platforms

-The MCP tools automatically filter advisories based on your platform.
+Platform metadata is preserved in advisory records and can be filtered by your policy layer.

 ## Security

@@ -260,7 +274,7 @@ Never manually edit the cache file - it will break signature verification.
 **Problem**: Advisory cache is empty or stale

 **Solution**:
-1. Check that `startAdvisoryCache()` is called in your host entry point
+1. Check that `AdvisoryCacheManager.initialize()` is called in your host entry point
 2. Verify network access to `clawsec.prompt.security`
 3. Check host logs for fetch errors
 4. Manually trigger: `curl https://clawsec.prompt.security/advisories/feed.json`
@@ -280,7 +294,7 @@ Never manually edit the cache file - it will break signature verification.
 **Problem**: Tools return errors about IPC

 **Solution**:
-1. Verify IPC handlers are registered in `host/ipc-handler.ts`
+1. Verify IPC handlers are registered in `src/ipc.ts`
 2. Check that IPC directory exists and is writable
 3. Ensure host process is running
 4. Check host logs for handler errors
@@ -290,8 +304,8 @@ Never manually edit the cache file - it will break signature verification.
 To remove ClawSec from NanoClaw:

 1. Remove MCP tool registration from `ipc-mcp-stdio.ts`
-2. Remove IPC handler registration from `host/ipc-handler.ts`
-3. Remove `startAdvisoryCache()` call from host entry point
+2. Remove IPC handler registration from `src/ipc.ts`
+3. Remove `AdvisoryCacheManager` initialization from host entry point
 4. Delete the skill directory: `rm -rf skills/clawsec-nanoclaw`
 5. Delete the cache file: `rm /workspace/project/data/clawsec-advisory-cache.json`
 6. Restart NanoClaw
@@ -56,9 +56,9 @@ ClawSec provides a complete security skill for NanoClaw deployments:
  - `clawsec_integrity_status` - View file baseline status
  - `clawsec_verify_audit` - Verify audit log hash chain

- **Advisory Cache Service**: Automatic feed fetching every 6 hours
+- **Advisory Cache Service**: Host-managed feed fetching with signature validation
 - **Signature Verification**: Ed25519-signed feeds ensure integrity
- **Platform Filtering**: Shows only relevant advisories for NanoClaw
+- **Exploitability Context**: Surfaces `exploitability_score` and rationale to reduce alert fatigue
 - **IPC Communication**: Container-safe host communication

 ### Installation
@@ -77,19 +77,19 @@ The skill integrates into three places:
 **1. MCP Tools** (container):
 ```typescript
 // container/agent-runner/src/ipc-mcp-stdio.ts
-import { clawsecTools } from '../../../skills/clawsec-nanoclaw/mcp-tools/advisory-tools.js';
+import '../../../skills/clawsec-nanoclaw/mcp-tools/advisory-tools.js';
 ```

 **2. IPC Handlers** (host):
 ```typescript
-// host/ipc-handler.ts
-import { registerClawSecHandlers } from '../skills/clawsec-nanoclaw/host-services/ipc-handlers.js';
+// src/ipc.ts
+import { handleAdvisoryIpc } from '../skills/clawsec-nanoclaw/host-services/ipc-handlers.js';
 ```

 **3. Cache Service** (host):
 ```typescript
-// host/index.ts
-import { startAdvisoryCache } from '../skills/clawsec-nanoclaw/host-services/advisory-cache.js';
+// src/index.ts
+import { AdvisoryCacheManager } from '../skills/clawsec-nanoclaw/host-services/advisory-cache.js';
 ```

 ### Advisory Feed
@@ -148,4 +148,4 @@ Planned features for future releases:

 - **Issues**: https://github.com/prompt-security/clawsec/issues
 - **Security**: security@prompt.security
- NanoClaw Repository: (link TBD)
+- NanoClaw Repository: https://github.com/qwibitai/nanoclaw
@@ -1,6 +1,6 @@
 ---
 name: clawsec-nanoclaw
-version: 0.0.1
+version: 0.0.2
 description: Use when checking for security vulnerabilities in NanoClaw skills, before installing new skills, or when asked about security advisories affecting the bot
 ---

@@ -10,7 +10,7 @@ Security advisory monitoring that protects your WhatsApp bot from known vulnerab

 ## Overview

-ClawSec provides MCP tools that check installed skills against a curated feed of security advisories. It prevents installation of vulnerable skills and alerts you to issues in existing ones.
+ClawSec provides MCP tools that check installed skills against a curated feed of security advisories. It prevents installation of vulnerable skills, includes exploitability context for triage, and alerts you to issues in existing ones.

 **Core principle:** Check before you install. Monitor what's running.

@@ -36,7 +36,7 @@ Do NOT use for:
 // Before installing any skill
 const safety = await tools.clawsec_check_skill_safety({
  skillName: 'new-skill',
-  version: '1.0.0'  // optional
+  skillVersion: '1.0.0'  // optional
 });

 if (!safety.safe) {
@@ -48,14 +48,16 @@ if (!safety.safe) {
 ### Security Audit

 ```typescript
-// Check all installed skills
+// Check all installed skills (defaults to ~/.claude/skills in the container)
 const result = await tools.clawsec_check_advisories({
-  skillsRoot: '/workspace/project/skills'  // optional
+  installRoot: '/home/node/.claude/skills'  // optional
 });

-if (result.criticalCount > 0) {
+if (result.matches.some((m) =>
+  m.advisory.severity === 'critical' || m.advisory.exploitability_score === 'high'
+)) {
  // Alert user immediately
-  console.error('CRITICAL vulnerabilities found!');
+  console.error('Urgent advisories found!');
 }
 ```

@@ -64,8 +66,8 @@ if (result.criticalCount > 0) {
 ```typescript
 // List advisories with filters
 const advisories = await tools.clawsec_list_advisories({
-  platform: 'nanoclaw',    // optional: nanoclaw, openclaw, or both
-  severity: 'critical'     // optional: critical, high, medium, low
+  severity: 'high',               // optional
+  exploitabilityScore: 'high'     // optional
 });
 ```

@@ -75,7 +77,7 @@ const advisories = await tools.clawsec_list_advisories({
 |------|------|---------------|
 | Pre-install check | `clawsec_check_skill_safety` | `skillName` |
 | Audit all skills | `clawsec_check_advisories` | `installRoot` (optional) |
-| Browse feed | `clawsec_list_advisories` | `severity`, `type` (optional) |
+| Browse feed | `clawsec_list_advisories` | `severity`, `type`, `exploitabilityScore` (optional) |
 | Verify package signature | `clawsec_verify_skill_package` | `packagePath` |
 | Refresh advisory cache | `clawsec_refresh_cache` | (none) |
 | Check file integrity | `clawsec_check_integrity` | `mode`, `autoRestore` (optional) |
@@ -110,7 +112,7 @@ if (safety.safe) {
 ```typescript
 // Add to scheduled tasks
 schedule_task({
-  prompt: "Check for security advisories using clawsec_check_advisories and alert if any critical issues found",
+  prompt: "Check advisories using clawsec_check_advisories and alert when critical or high-exploitability matches appear",
  schedule_type: "cron",
  schedule_value: "0 9 * * *"  // Daily at 9am
 });
@@ -125,8 +127,8 @@ You: I'll check installed skills for known vulnerabilities.
 [Use clawsec_check_advisories]

 Response:
-✅ No critical issues found.
- 2 low-severity advisories (not urgent)
+✅ No urgent issues found.
+- 2 low-severity/low-exploitability advisories
 - All skills up to date
 ```

@@ -146,30 +148,33 @@ const safety = await tools.clawsec_check_skill_safety({
 if (safety.safe) await installSkill('untrusted-skill');
 ```

-### ❌ Ignoring platform filters
+### ❌ Ignoring exploitability context
 ```typescript
-// DON'T: Check OpenClaw advisories on NanoClaw
-const advisories = await tools.clawsec_list_advisories({
-  platform: 'openclaw'  // Wrong platform!
-});
+// DON'T: Use severity only
+if (advisory.severity === 'high') {
+  notifyNow(advisory);
+}
 ```

 ```typescript
-// DO: Use correct platform or let it auto-filter
-const advisories = await tools.clawsec_list_advisories({
-  platform: 'nanoclaw'  // Correct
-});
+// DO: Use exploitability + severity
+if (
+  advisory.exploitability_score === 'high' ||
+  advisory.severity === 'critical'
+) {
+  notifyNow(advisory);
+}
 ```

 ### ❌ Skipping critical severity
 ```typescript
-// DON'T: Only check low severity
-if (result.lowCount > 0) alert();
+// DON'T: Ignore high exploitability in medium severity advisories
+if (advisory.severity === 'critical') alert();
 ```

 ```typescript
-// DO: Prioritize critical and high
-if (result.criticalCount > 0 || result.highCount > 0) {
+// DO: Prioritize exploitability and severity together
+if (advisory.exploitability_score === 'high' || advisory.severity === 'critical') {
  // Alert immediately
 }
 ```
@@ -182,7 +187,7 @@ if (result.criticalCount > 0 || result.highCount > 0) {

 **Signature Verification**: Ed25519 signed feeds

-**Cache Location**: `/workspace/project/data/clawsec-cache.json`
+**Cache Location**: `/workspace/project/data/clawsec-advisory-cache.json`

 See [INSTALL.md](./INSTALL.md) for setup and [docs/](./docs/) for advanced usage.

@@ -16,6 +16,7 @@ import crypto from 'node:crypto';
 import fs from 'node:fs/promises';
 import https from 'node:https';
 import path from 'node:path';
+import { evaluateAdvisoryRisk } from '../lib/risk.js';

 // ClawSec public key (from clawsec-signing-public.pem)
 const PUBLIC_KEY_PEM = `-----BEGIN PUBLIC KEY-----
@@ -35,6 +36,8 @@ export interface Advisory {
  action?: string;
  published?: string;
  updated?: string;
+  exploitability_score?: 'high' | 'medium' | 'low' | 'unknown' | string;
+  exploitability_rationale?: string;
  affected: string[];
 }

@@ -376,42 +379,5 @@ export function evaluateSkillSafety(advisories: Advisory[]): {
  recommendation: 'install' | 'block' | 'review';
  reason: string;
 } {
-  if (advisories.length === 0) {
-    return { safe: true, recommendation: 'install', reason: 'No advisories found' };
-  }
-
-  const hasMalicious = advisories.some((a) => a.type === 'malicious');
-  const hasRemoveAction = advisories.some((a) => a.action === 'remove');
-  const hasCritical = advisories.some((a) => a.severity === 'critical');
-  const hasHigh = advisories.some((a) => a.severity === 'high');
-
-  if (hasMalicious || hasRemoveAction) {
-    return {
-      safe: false,
-      recommendation: 'block',
-      reason: 'Malicious skill or removal recommended',
-    };
-  }
-
-  if (hasCritical) {
-    return {
-      safe: false,
-      recommendation: 'block',
-      reason: 'Critical security advisory',
-    };
-  }
-
-  if (hasHigh) {
-    return {
-      safe: false,
-      recommendation: 'review',
-      reason: 'High severity advisory - user review recommended',
-    };
-  }
-
-  return {
-    safe: false,
-    recommendation: 'review',
-    reason: 'Advisory found - review before installing',
-  };
+  return evaluateAdvisoryRisk(advisories);
 }
@@ -121,6 +121,34 @@ export function versionMatches(version: string, versionSpec: string): boolean {
  return false;
 }

+/**
+ * Checks whether an affected specifier matches a skill name/version.
+ * Optionally matches against a skill directory name as alias.
+ */
+export function matchesAffectedSpecifier(
+  affected: string,
+  skillName: string,
+  skillVersion: string | null,
+  skillDirName?: string
+): boolean {
+  const parsed = parseAffectedSpecifier(affected);
+  if (!parsed) return false;
+
+  const normalizedTarget = normalizeSkillName(parsed.name);
+  const normalizedSkillName = normalizeSkillName(skillName);
+  const normalizedDirName = skillDirName ? normalizeSkillName(skillDirName) : null;
+
+  if (normalizedTarget !== normalizedSkillName && normalizedTarget !== normalizedDirName) {
+    return false;
+  }
+
+  if (!skillVersion) {
+    return true;
+  }
+
+  return versionMatches(skillVersion, parsed.versionSpec);
+}
+
 /**
 * Loads advisory feed from a remote URL with signature verification.
 */
@@ -269,10 +297,12 @@ export async function loadFeed(
 export function advisoryLooksHighRisk(advisory: Advisory): boolean {
  const type = advisory.type.toLowerCase();
  const severity = advisory.severity.toLowerCase();
+  const exploitability = (advisory.exploitability_score || 'unknown').toLowerCase();
  const combined = `${advisory.title} ${advisory.description} ${advisory.action}`.toLowerCase();

-  if (type === 'malicious_skill' || type === 'malicious_plugin') return true;
+  if (type.includes('malicious')) return true;
  if (severity === 'critical') return true;
+  if (exploitability === 'high') return true;
  if (/\b(malicious|exfiltrate|exfiltration|backdoor|trojan|stealer|credential theft)\b/.test(combined)) return true;
  if (/\b(remove|uninstall|disable|do not use|quarantine)\b/.test(combined)) return true;

@@ -294,15 +324,7 @@ export function findAdvisoryMatches(
    if (affected.length === 0) continue;

    for (const specifier of affected) {
-      const parsed = parseAffectedSpecifier(specifier);
-      if (!parsed) continue;
-
-      if (normalizeSkillName(parsed.name) !== normalizeSkillName(skillName)) {
-        continue;
-      }
-
-      // If version specified, check if it matches
-      if (version && !versionMatches(version, parsed.versionSpec)) {
+      if (!matchesAffectedSpecifier(specifier, skillName, version)) {
        continue;
      }

@@ -0,0 +1,88 @@
+/**
+ * Shared advisory risk evaluation for NanoClaw host + MCP layers.
+ */
+
+export type SkillSafetyRecommendation = 'install' | 'block' | 'review';
+
+export interface AdvisoryRiskInput {
+  severity?: string;
+  type?: string;
+  action?: string;
+  exploitability_score?: string;
+}
+
+export interface AdvisoryRiskEvaluation {
+  safe: boolean;
+  recommendation: SkillSafetyRecommendation;
+  reason: string;
+}
+
+export function normalizeExploitabilityScore(score: unknown): 'high' | 'medium' | 'low' | 'unknown' {
+  const value = String(score || '').toLowerCase().trim();
+  if (value === 'high' || value === 'medium' || value === 'low') {
+    return value;
+  }
+  return 'unknown';
+}
+
+export function evaluateAdvisoryRisk(advisories: AdvisoryRiskInput[]): AdvisoryRiskEvaluation {
+  if (advisories.length === 0) {
+    return { safe: true, recommendation: 'install', reason: 'No advisories found' };
+  }
+
+  const hasMalicious = advisories.some((a) => String(a.type || '').toLowerCase().includes('malicious'));
+  const hasRemoveAction = advisories.some((a) =>
+    /\b(remove|uninstall|disable|quarantine|block)\b/i.test(String(a.action || ''))
+  );
+  const hasCritical = advisories.some((a) => String(a.severity || '').toLowerCase() === 'critical');
+  const hasHigh = advisories.some((a) => String(a.severity || '').toLowerCase() === 'high');
+  const hasHighExploitability = advisories.some(
+    (a) => normalizeExploitabilityScore(a.exploitability_score) === 'high'
+  );
+
+  if (hasMalicious || hasRemoveAction) {
+    return {
+      safe: false,
+      recommendation: 'block',
+      reason: 'Malicious skill or removal recommended by ClawSec',
+    };
+  }
+
+  if (hasCritical && hasHighExploitability) {
+    return {
+      safe: false,
+      recommendation: 'block',
+      reason: 'Critical advisory with high exploitability context - do not install',
+    };
+  }
+
+  if (hasCritical) {
+    return {
+      safe: false,
+      recommendation: 'block',
+      reason: 'Critical security advisory - do not install',
+    };
+  }
+
+  if (hasHighExploitability) {
+    return {
+      safe: false,
+      recommendation: 'review',
+      reason: 'High exploitability advisory - urgent user review strongly recommended',
+    };
+  }
+
+  if (hasHigh) {
+    return {
+      safe: false,
+      recommendation: 'review',
+      reason: 'High severity advisory - user review strongly recommended',
+    };
+  }
+
+  return {
+    safe: false,
+    recommendation: 'review',
+    reason: 'Advisory found - review details before installing',
+  };
+}
@@ -15,6 +15,8 @@ export interface Advisory {
  references: string[];
  cvss_score?: number;
  nvd_url?: string;
+  exploitability_score?: 'high' | 'medium' | 'low' | 'unknown';
+  exploitability_rationale?: string;
  source?: string;
  github_issue_url?: string;
  reporter?: {
@@ -11,6 +11,8 @@
 import fs from 'fs';
 import path from 'path';
 import { z } from 'zod';
+import { evaluateAdvisoryRisk, normalizeExploitabilityScore } from '../lib/risk.js';
+import { matchesAffectedSpecifier } from '../lib/advisories.js';

 // These variables are provided by the host environment (ipc-mcp-stdio.ts)
 // when this code is integrated into the NanoClaw container agent.
@@ -18,8 +20,10 @@ declare const server: { tool: (...args: any[]) => void };
 declare function writeIpcFile(dir: string, data: any): void;
 declare const TASKS_DIR: string;
 declare const groupFolder: string;
+const CACHE_FILE = '/workspace/project/data/clawsec-advisory-cache.json';

-// Add these helper functions to the file:
+const severityOrder: Record<string, number> = { critical: 0, high: 1, medium: 2, low: 3 };
+const exploitabilityOrder: Record<string, number> = { high: 0, medium: 1, low: 2, unknown: 3 };

 /**
 * Discover installed skills in a directory
@@ -84,10 +88,7 @@ function findAdvisoryMatches(
      const matchedAffected: string[] = [];

      for (const affected of advisory.affected || []) {
-        const atIndex = affected.lastIndexOf('@');
-        const affectedName = atIndex > 0 ? affected.slice(0, atIndex) : affected;
-
-        if (affectedName === skill.name || affectedName === skill.dirName) {
+        if (matchesAffectedSpecifier(affected, skill.name, skill.version, skill.dirName)) {
          matchedAffected.push(affected);
        }
      }
@@ -123,10 +124,8 @@ server.tool(
    }

    // Read cache from shared mount
-    const cacheFile = '/workspace/project/data/clawsec-advisory-cache.json';
-
    try {
-      const cacheData = JSON.parse(fs.readFileSync(cacheFile, 'utf8'));
+      const cacheData = JSON.parse(fs.readFileSync(CACHE_FILE, 'utf8'));
      const installRoot = args.installRoot || path.join(process.env.HOME || '~', '.claude', 'skills');

      // Discover installed skills
@@ -153,6 +152,8 @@ server.tool(
            description: m.advisory.description,
            action: m.advisory.action,
            published: m.advisory.published,
+            exploitability_score: normalizeExploitabilityScore(m.advisory.exploitability_score),
+            exploitability_rationale: m.advisory.exploitability_rationale || null,
          },
          skill: m.skill,
          matchedAffected: m.matchedAffected,
@@ -187,17 +188,13 @@ server.tool(
    skillVersion: z.string().optional().describe('Version of skill (optional, for version-specific checks)'),
  },
  async (args) => {
-    const cacheFile = '/workspace/project/data/clawsec-advisory-cache.json';
-
    try {
-      const cacheData = JSON.parse(fs.readFileSync(cacheFile, 'utf8'));
+      const cacheData = JSON.parse(fs.readFileSync(CACHE_FILE, 'utf8'));

      // Find matching advisories for this skill
      const matchingAdvisories = cacheData.feed.advisories.filter((advisory: any) =>
        advisory.affected.some((affected: string) => {
-          const atIndex = affected.lastIndexOf('@');
-          const affectedName = atIndex > 0 ? affected.slice(0, atIndex) : affected;
-          return affectedName === args.skillName;
+          return matchesAffectedSpecifier(affected, args.skillName, args.skillVersion || null);
        })
      );

@@ -215,34 +212,13 @@ server.tool(
        };
      }

-      // Evaluate severity
-      const hasMalicious = matchingAdvisories.some((a: any) => a.type === 'malicious');
-      const hasRemoveAction = matchingAdvisories.some((a: any) => a.action === 'remove');
-      const hasCritical = matchingAdvisories.some((a: any) => a.severity === 'critical');
-      const hasHigh = matchingAdvisories.some((a: any) => a.severity === 'high');
-
-      let recommendation: 'install' | 'block' | 'review';
-      let reason: string;
-
-      if (hasMalicious || hasRemoveAction) {
-        recommendation = 'block';
-        reason = 'Malicious skill or removal recommended by ClawSec';
-      } else if (hasCritical) {
-        recommendation = 'block';
-        reason = 'Critical security advisory - do not install';
-      } else if (hasHigh) {
-        recommendation = 'review';
-        reason = 'High severity advisory - user review strongly recommended';
-      } else {
-        recommendation = 'review';
-        reason = 'Advisory found - review details before installing';
-      }
+      const risk = evaluateAdvisoryRisk(matchingAdvisories);

      return {
        content: [{
          type: 'text' as const,
          text: JSON.stringify({
-            safe: false, // Always false when advisories exist
+            safe: risk.safe,
            advisories: matchingAdvisories.map((a: any) => ({
              id: a.id,
              severity: a.severity,
@@ -252,10 +228,13 @@ server.tool(
              action: a.action,
              published: a.published,
              affected: a.affected,
+              exploitability_score: normalizeExploitabilityScore(a.exploitability_score),
+              exploitability_rationale: a.exploitability_rationale || null,
            })),
-            recommendation,
-            reason,
+            recommendation: risk.recommendation,
+            reason: risk.reason,
            skillName: args.skillName,
+            skillVersion: args.skillVersion || null,
            advisoryCount: matchingAdvisories.length,
          }, null, 2),
        }],
@@ -280,18 +259,18 @@ server.tool(

 server.tool(
  'clawsec_list_advisories',
-  'List ClawSec advisories with optional filtering. Use this to browse security advisories, filter by severity/type, or search for specific affected skills.',
+  'List ClawSec advisories with optional filtering. Use this to browse security advisories, filter by severity/type/exploitability, or search for specific affected skills.',
  {
    severity: z.enum(['critical', 'high', 'medium', 'low']).optional().describe('Filter by severity level'),
-    type: z.enum(['vulnerability', 'malicious', 'deprecated']).optional().describe('Filter by advisory type'),
+    type: z.string().optional().describe('Filter by advisory type (for example: vulnerable_skill, malicious_skill, prompt_injection)'),
+    exploitabilityScore: z.enum(['high', 'medium', 'low', 'unknown']).optional()
+      .describe('Filter by exploitability score'),
    affectedSkill: z.string().optional().describe('Filter by affected skill name (partial match supported)'),
    limit: z.number().optional().describe('Maximum number of results (default: unlimited)'),
  },
  async (args) => {
-    const cacheFile = '/workspace/project/data/clawsec-advisory-cache.json';
-
    try {
-      const cacheData = JSON.parse(fs.readFileSync(cacheFile, 'utf8'));
+      const cacheData = JSON.parse(fs.readFileSync(CACHE_FILE, 'utf8'));
      let advisories = [...cacheData.feed.advisories];

      // Apply filters
@@ -299,7 +278,13 @@ server.tool(
        advisories = advisories.filter((a: any) => a.severity === args.severity);
      }
      if (args.type) {
-        advisories = advisories.filter((a: any) => a.type === args.type);
+        const typeFilter = String(args.type).toLowerCase().trim();
+        advisories = advisories.filter((a: any) => String(a.type || '').toLowerCase().trim() === typeFilter);
+      }
+      if (args.exploitabilityScore) {
+        advisories = advisories.filter(
+          (a: any) => normalizeExploitabilityScore(a.exploitability_score) === args.exploitabilityScore
+        );
      }
      if (args.affectedSkill) {
        advisories = advisories.filter((a: any) =>
@@ -307,9 +292,13 @@ server.tool(
        );
      }

-      // Sort by severity (critical first) and published date (newest first)
-      const severityOrder: Record<string, number> = { critical: 0, high: 1, medium: 2, low: 3 };
+      // Sort by exploitability first, then severity, then publish date (newest first).
      advisories.sort((a: any, b: any) => {
+        const exploitabilityDiff =
+          (exploitabilityOrder[normalizeExploitabilityScore(a.exploitability_score)] ?? 999) -
+          (exploitabilityOrder[normalizeExploitabilityScore(b.exploitability_score)] ?? 999);
+        if (exploitabilityDiff !== 0) return exploitabilityDiff;
+
        const severityDiff = (severityOrder[a.severity] || 999) - (severityOrder[b.severity] || 999);
        if (severityDiff !== 0) return severityDiff;
        return (b.published || '').localeCompare(a.published || '');
@@ -336,6 +325,8 @@ server.tool(
              action: a.action,
              published: a.published,
              affected: a.affected,
+              exploitability_score: normalizeExploitabilityScore(a.exploitability_score),
+              exploitability_rationale: a.exploitability_rationale || null,
            })),
            total: cacheData.feed.advisories.length,
            filtered: originalCount,
@@ -343,6 +334,7 @@ server.tool(
            filters: {
              severity: args.severity || null,
              type: args.type || null,
+              exploitabilityScore: args.exploitabilityScore || null,
              affectedSkill: args.affectedSkill || null,
              limit: args.limit || null,
            },
@@ -1,6 +1,6 @@
 {
  "name": "clawsec-nanoclaw",
-  "version": "0.0.1",
+  "version": "0.0.2",
  "description": "ClawSec security suite for NanoClaw - Advisory feed monitoring, MCP tools for vulnerability checking, and Ed25519 signature verification for containerized WhatsApp bot agents",
  "author": "prompt-security",
  "license": "AGPL-3.0-or-later",
@@ -27,6 +27,11 @@
        "required": true,
        "description": "NanoClaw skill documentation"
      },
+      {
+        "path": "CHANGELOG.md",
+        "required": true,
+        "description": "Version history and release notes"
+      },
      {
        "path": "INSTALL.md",
        "required": true,
@@ -62,6 +67,11 @@
        "required": true,
        "description": "TypeScript type definitions"
      },
+      {
+        "path": "lib/risk.ts",
+        "required": true,
+        "description": "Shared advisory risk evaluation logic for host and MCP tools"
+      },
      {
        "path": "advisories/feed-signing-public.pem",
        "required": true,
@@ -112,9 +122,10 @@
  "capabilities": [
    "Advisory feed monitoring from clawsec.prompt.security",
    "MCP tools for agent-initiated vulnerability scans",
+    "Exploitability-aware advisory prioritization for agent environments",
    "Pre-installation skill safety checks",
    "Ed25519 signature verification for advisory feeds",
-    "Platform-specific advisory filtering (nanoclaw vs openclaw)",
+    "Platform metadata preserved in advisory records for downstream filtering",
    "Containerized agent support with IPC communication"
  ],
  "nanoclaw": {
@@ -135,7 +146,7 @@
    },
    "integration": {
      "mcp_tools_file": "container/agent-runner/src/ipc-mcp-stdio.ts",
-      "ipc_handlers_file": "host/ipc-handler.ts",
+      "ipc_handlers_file": "src/ipc.ts",
      "cache_location": "/workspace/project/data/clawsec-advisory-cache.json"
    }
  }
@@ -5,6 +5,21 @@ All notable changes to the ClawSec Suite will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

+## [0.1.4] - 2026-02-28
+
+### Added
+
+- Advisory output snippets now include exploitability context in suite quick-check and heartbeat examples.
+
+### Changed
+
+- Clarified exploitability guidance to match runtime score values (`high|medium|low|unknown`).
+- Prioritization guidance now emphasizes high-exploitability advisories for immediate handling.
+
+### Fixed
+
+- Kept exploitability enrichment in advisory workflows non-fatal per item so a single analysis failure does not abort feed updates.
+
 ## [0.1.3]

 ### Added
@@ -121,6 +121,7 @@ else
    while IFS= read -r id; do
      [ -z "$id" ] && continue
      jq -r --arg id "$id" '.advisories[] | select(.id == $id) | "- [\(.severity | ascii_upcase)] \(.id): \(.title)"' "$FEED_TMP"
+      jq -r --arg id "$id" '.advisories[] | select(.id == $id) | "  Exploitability: \(.exploitability_score // "unknown" | ascii_upcase)"' "$FEED_TMP"
      jq -r --arg id "$id" '.advisories[] | select(.id == $id) | "  Action: \(.action // "Review advisory details")"' "$FEED_TMP"
    done < "$NEW_IDS_FILE"
  else
@@ -194,8 +195,18 @@ fi
 Heartbeat output should include:
 - suite version status,
 - advisory feed status,
- new advisory list (if any),
+- new advisory list (if any) with exploitability scores,
 - installed skills that appear in advisory `affected` lists,
 - and a double-confirmation reminder before risky install/remove actions.

-If your runtime sends alerts, treat `critical` and `high` advisories affecting installed skills as immediate notifications.
+### Exploitability-Based Prioritization
+
+When alerting on advisories, prioritize by **exploitability score** in addition to severity:
+
+- `high` exploitability: Trivially or easily exploitable with public tooling, immediate action required
+- `medium` exploitability: Exploitable with specific conditions, standard priority
+- `low` exploitability: Difficult to exploit or theoretical, low priority
+
+**Priority Rule**: A HIGH severity + HIGH exploitability CVE should be treated more urgently than a CRITICAL severity + LOW exploitability CVE.
+
+If your runtime sends alerts, treat `high` exploitability advisories affecting installed skills as immediate notifications, regardless of severity rating.
@@ -1,6 +1,6 @@
 ---
 name: clawsec-suite
-version: 0.1.3
+version: 0.1.4
 description: ClawSec suite manager with embedded advisory-feed monitoring, cryptographic signature verification, approval-gated malicious-skill response, and guided setup for additional security skills.
 homepage: https://clawsec.prompt.security
 clawdis:
@@ -234,12 +234,25 @@ if [ -s "$NEW_IDS_FILE" ]; then
  while IFS= read -r id; do
    [ -z "$id" ] && continue
    jq -r --arg id "$id" '.advisories[] | select(.id == $id) | "- [\(.severity | ascii_upcase)] \(.id): \(.title)"' "$TMP/feed.json"
+    jq -r --arg id "$id" '.advisories[] | select(.id == $id) | "  Exploitability: \(.exploitability_score // "unknown" | ascii_upcase)"' "$TMP/feed.json"
  done < "$NEW_IDS_FILE"
 else
  echo "FEED_OK - no new advisories"
 fi
 ```

+## Exploitability Context
+
+Advisories in the feed can include `exploitability_score` and `exploitability_rationale` fields to help agents prioritize real-world threats:
+
+- **Exploitability scores**: `high`, `medium`, `low`, or `unknown`
+- **Context-aware assessment**: Considers attack vector, authentication requirements, and AI agent deployment patterns
+- **Exploit availability**: Detects public exploits and weaponization status
+
+When processing advisories, prioritize by exploitability in addition to severity. A HIGH severity + HIGH exploitability CVE is more urgent than a CRITICAL severity + LOW exploitability CVE.
+
+For detailed methodology, see the [exploitability scoring documentation](../../wiki/exploitability-scoring.md).
+
 ## Heartbeat Integration

 Use the suite heartbeat script as the single periodic security check entrypoint:
@@ -1,6 +1,6 @@
 {
  "name": "clawsec-suite",
-  "version": "0.1.3",
+  "version": "0.1.4",
  "description": "ClawSec suite manager with embedded advisory-feed monitoring, cryptographic signature verification, approval-gated malicious-skill response, and guided setup for additional security skills.",
  "author": "prompt-security",
  "license": "AGPL-3.0-or-later",
@@ -0,0 +1,531 @@
+#!/usr/bin/env python3
+"""
+Exploitability Analyzer - Analyzes CVE exploitability in OpenClaw/NanoClaw deployments
+
+Usage:
+    python utils/analyze_exploitability.py --help
+    echo '{"cve_id":"CVE-2026-27488","cvss_score":7.3}' | python utils/analyze_exploitability.py --json
+    python utils/analyze_exploitability.py --test-cases
+
+Example:
+    cat cve-data.json | python utils/analyze_exploitability.py --json
+"""
+
+import argparse
+import json
+import sys
+from typing import Any
+
+
+def parse_cvss_vector(vector_string: str) -> dict[str, str]:
+    """
+    Parse CVSS v2, v3.0, or v3.1 vector string into components.
+
+    Args:
+        vector_string: CVSS vector (e.g., "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H")
+
+    Returns:
+        Dictionary of CVSS metrics and values
+    """
+    if not vector_string:
+        return {}
+
+    metrics = {}
+    normalized = vector_string.strip()
+
+    # Remove leading CVSS v3.x prefix if present (e.g., "CVSS:3.1/")
+    if normalized.startswith("CVSS:3"):
+        _, separator, remainder = normalized.partition("/")
+        normalized = remainder if separator else ""
+
+    # Remove surrounding parentheses/whitespace used by some CVSS v2 strings.
+    normalized = normalized.strip().strip("()").strip()
+    if not normalized:
+        return metrics
+
+    # Parse all vector formats with shared key/value extraction logic.
+    for part in normalized.split("/"):
+        if ":" in part:
+            key, value = part.split(":", 1)
+            metrics[key] = value
+
+    return metrics
+
+
+def analyze_attack_vector(cvss_metrics: dict[str, str]) -> dict[str, Any]:
+    """
+    Analyze attack vector from CVSS metrics.
+
+    Args:
+        cvss_metrics: Parsed CVSS metrics dictionary
+
+    Returns:
+        Dictionary with attack vector analysis
+    """
+    analysis = {
+        "is_network_accessible": False,
+        "requires_authentication": True,
+        "requires_user_interaction": True,
+        "complexity": "unknown"
+    }
+
+    # Attack Vector (AV)
+    av = cvss_metrics.get("AV", "")
+    if av == "N":  # Network
+        analysis["is_network_accessible"] = True
+    elif av == "A":  # Adjacent Network
+        analysis["is_network_accessible"] = True
+    elif av in ["L", "P"]:  # Local or Physical
+        analysis["is_network_accessible"] = False
+
+    # Privileges Required (PR) / Authentication (AU for v2)
+    pr = cvss_metrics.get("PR", cvss_metrics.get("Au", ""))
+    if pr in ["N", "NONE"]:
+        analysis["requires_authentication"] = False
+    elif pr in ["L", "H", "SINGLE", "MULTIPLE"]:
+        analysis["requires_authentication"] = True
+
+    # User Interaction (UI)
+    ui = cvss_metrics.get("UI", "")
+    if ui == "N":  # None
+        analysis["requires_user_interaction"] = False
+    elif ui == "R":  # Required
+        analysis["requires_user_interaction"] = True
+
+    # Attack Complexity (AC)
+    ac = cvss_metrics.get("AC", "")
+    if ac == "L":
+        analysis["complexity"] = "low"
+    elif ac in ["M", "H"]:
+        analysis["complexity"] = "high"
+
+    return analysis
+
+
+def detect_exploit_availability(references: list[str]) -> dict[str, Any]:
+    """
+    Detect if exploits are publicly available based on reference URLs.
+
+    Args:
+        references: List of reference URLs
+
+    Returns:
+        Dictionary with exploit_available (bool) and exploit_sources (list)
+    """
+    exploit_indicators = [
+        "exploit-db.com",
+        "exploit-database",
+        "exploitdb",
+        "packetstormsecurity.com",
+        "packetstorm",
+        "github.com/exploit",
+        "github.com/poc",
+        "github.com/proof-of-concept",
+        "metasploit",
+        "exploit/",
+        "/exploit",
+        "/poc",
+        "/proof-of-concept",
+        "exploitability",
+        "exploit-code",
+    ]
+
+    exploit_sources = []
+    for ref in references:
+        ref_lower = ref.lower()
+        for indicator in exploit_indicators:
+            if indicator in ref_lower:
+                exploit_sources.append(ref)
+                break
+
+    return {
+        "exploit_available": len(exploit_sources) > 0,
+        "exploit_sources": exploit_sources
+    }
+
+
+def analyze_exploitability(cve_data: dict[str, Any], check_exploits: bool = False) -> dict[str, Any]:
+    """
+    Analyze CVE exploitability for OpenClaw/NanoClaw deployments.
+
+    Args:
+        cve_data: Dictionary containing CVE information with keys:
+                  - cve_id: CVE identifier
+                  - cvss_score: CVSS base score (float)
+                  - cvss_vector: CVSS vector string (optional)
+                  - type: Vulnerability type
+                  - description: CVE description text
+                  - references: List of reference URLs (optional)
+        check_exploits: Whether to check references for exploit availability
+
+    Returns:
+        Dictionary with exploitability_score (high/medium/low/unknown) and rationale
+    """
+    cve_id = cve_data.get("cve_id", "unknown")
+    cvss_score = cve_data.get("cvss_score", 0.0)
+    cvss_vector = cve_data.get("cvss_vector", "")
+    vuln_type = cve_data.get("type", "")
+    description = cve_data.get("description", "")
+    references = cve_data.get("references", [])
+
+    # Parse CVSS vector if available
+    cvss_metrics = parse_cvss_vector(cvss_vector)
+    attack_analysis = analyze_attack_vector(cvss_metrics)
+
+    # Initial scoring based on CVSS
+    score = "unknown"
+    rationale_parts = []
+
+    # CVSS-based baseline
+    if cvss_score >= 9.0:
+        score = "high"
+        rationale_parts.append(f"Critical CVSS score ({cvss_score})")
+    elif cvss_score >= 7.0:
+        score = "high"
+        rationale_parts.append(f"High CVSS score ({cvss_score})")
+    elif cvss_score >= 4.0:
+        score = "medium"
+        rationale_parts.append(f"Medium CVSS score ({cvss_score})")
+    elif cvss_score > 0:
+        score = "low"
+        rationale_parts.append(f"Low CVSS score ({cvss_score})")
+    else:
+        score = "unknown"
+        rationale_parts.append("No CVSS score available")
+
+    # Adjust based on attack vector analysis
+    if attack_analysis["is_network_accessible"]:
+        if not attack_analysis["requires_authentication"] and not attack_analysis["requires_user_interaction"]:
+            # Network accessible, no auth, no user interaction = highly exploitable
+            if score == "medium":
+                score = "high"
+            rationale_parts.append("remotely exploitable without authentication")
+        else:
+            rationale_parts.append("network accessible")
+    else:
+        # Local-only vulnerabilities are less critical in agent deployments
+        if score == "high":
+            score = "medium"
+        rationale_parts.append("requires local access")
+
+    # OpenClaw/NanoClaw deployment context - adjust based on vulnerability type
+    vuln_type_lower = vuln_type.lower()
+    description_lower = description.lower()
+
+    # High-risk vulnerability types in AI agent deployments
+    if any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "ssrf", "server_side_request_forgery", "server-side request forgery"
+    ]):
+        # SSRF is critical for agents that make external API calls
+        if score != "high" and cvss_score >= 6.0:
+            score = "high"
+        rationale_parts.append("SSRF affects agents making external requests")
+
+    elif any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "path_traversal", "path traversal", "directory traversal", "file_inclusion"
+    ]):
+        # Path traversal is critical for agents with file system access
+        if score != "high" and cvss_score >= 6.0:
+            score = "high"
+        rationale_parts.append("path traversal affects agents with file access")
+
+    elif any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "rce", "remote_code_execution", "remote code execution", "code_injection",
+        "command_injection", "command injection", "arbitrary code"
+    ]):
+        # RCE is always critical regardless of other factors
+        score = "high"
+        rationale_parts.append("RCE is critical in agent deployments")
+
+    elif any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "prototype_pollution", "prototype pollution"
+    ]):
+        # Prototype pollution in Node.js agents can lead to RCE
+        if score == "low":
+            score = "medium"
+        rationale_parts.append("prototype pollution can escalate in Node.js agents")
+
+    elif any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "xss", "cross_site_scripting", "cross-site scripting", "reflected xss", "stored xss"
+    ]):
+        # XSS is lower risk in headless agent deployments (no browser rendering)
+        if score == "high" and not attack_analysis["is_network_accessible"]:
+            score = "medium"
+        rationale_parts.append("XSS has limited impact in headless agents")
+
+    elif any(keyword in vuln_type_lower or keyword in description_lower for keyword in [
+        "sql_injection", "sql injection", "nosql injection"
+    ]):
+        # SQL injection depends on whether agent uses databases
+        if attack_analysis["is_network_accessible"] and not attack_analysis["requires_authentication"]:
+            if score == "medium":
+                score = "high"
+        rationale_parts.append("injection affects agents with database access")
+
+    # Check for exploit availability if requested
+    exploit_info = {"exploit_available": False, "exploit_sources": []}
+    if check_exploits and references:
+        exploit_info = detect_exploit_availability(references)
+        if exploit_info["exploit_available"]:
+            # Elevate score if public exploits exist
+            if score == "low":
+                score = "medium"
+            elif score == "medium":
+                score = "high"
+            elif score == "unknown" and cvss_score > 0:
+                # If we have some CVSS score but it was unknown, upgrade to at least medium
+                score = "medium"
+
+            exploit_count = len(exploit_info["exploit_sources"])
+            source_suffix = "s" if exploit_count > 1 else ""
+            rationale_parts.append(
+                f"public exploit available ({exploit_count} source{source_suffix})"
+            )
+
+    # Build rationale string
+    rationale = "; ".join(rationale_parts[:5])  # Limit to first 5 parts for context
+
+    result = {
+        "cve_id": cve_id,
+        "exploitability_score": score,
+        "exploitability_rationale": rationale,
+        "attack_vector_analysis": attack_analysis
+    }
+
+    # Include exploit info if check_exploits was enabled
+    if check_exploits:
+        result["exploit_detection"] = exploit_info
+
+    return result
+
+
+def run_test_cases():
+    """
+    Run comprehensive test cases for attack vector analysis.
+    Tests CVSS vector parsing and attack vector analysis logic.
+    """
+    test_cases = [
+        {
+            "name": "CVSS 3.1 - Network accessible, no auth, no UI (critical)",
+            "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": False,
+                "requires_user_interaction": False,
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS 3.1 - Network accessible, requires auth",
+            "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": True,
+                "requires_user_interaction": False,
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS 3.1 - Network accessible, requires UI",
+            "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": False,
+                "requires_user_interaction": True,
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS 3.1 - Local access required",
+            "cvss_vector": "CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
+            "expected": {
+                "is_network_accessible": False,
+                "requires_authentication": False,
+                "requires_user_interaction": False,
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS 3.1 - Adjacent network, high auth",
+            "cvss_vector": "CVSS:3.1/AV:A/AC:H/PR:H/UI:R/S:U/C:L/I:L/A:L",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": True,
+                "requires_user_interaction": True,
+                "complexity": "high"
+            }
+        },
+        {
+            "name": "CVSS 3.0 - Physical access required",
+            "cvss_vector": "CVSS:3.0/AV:P/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
+            "expected": {
+                "is_network_accessible": False,
+                "requires_authentication": False,
+                "requires_user_interaction": False,
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS v2 - Network, no auth required",
+            "cvss_vector": "(AV:N/AC:L/Au:N/C:C/I:C/A:C)",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": False,
+                "requires_user_interaction": True,  # v2 doesn't have UI, defaults to True
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "CVSS v2 - Network, single auth",
+            "cvss_vector": "AV:N/AC:M/Au:SINGLE/C:P/I:P/A:P",
+            "expected": {
+                "is_network_accessible": True,
+                "requires_authentication": True,
+                "requires_user_interaction": True,  # v2 doesn't have UI, defaults to True
+                "complexity": "high"
+            }
+        },
+        {
+            "name": "CVSS v2 - Local access, multiple auth",
+            "cvss_vector": "(AV:L/AC:L/Au:MULTIPLE/C:C/I:C/A:C)",
+            "expected": {
+                "is_network_accessible": False,
+                "requires_authentication": True,
+                "requires_user_interaction": True,  # v2 doesn't have UI, defaults to True
+                "complexity": "low"
+            }
+        },
+        {
+            "name": "Empty CVSS vector",
+            "cvss_vector": "",
+            "expected": {
+                "is_network_accessible": False,
+                "requires_authentication": True,
+                "requires_user_interaction": True,
+                "complexity": "unknown"
+            }
+        }
+    ]
+
+    print("Running attack vector analysis test cases...")
+    print("=" * 70)
+
+    passed = 0
+    failed = 0
+
+    for i, test in enumerate(test_cases, 1):
+        print(f"\nTest {i}/{len(test_cases)}: {test['name']}")
+        print(f"  CVSS Vector: {test['cvss_vector']}")
+
+        # Parse CVSS vector and analyze attack vector
+        cvss_metrics = parse_cvss_vector(test['cvss_vector'])
+        result = analyze_attack_vector(cvss_metrics)
+
+        # Compare with expected results
+        test_passed = True
+        for key, expected_value in test['expected'].items():
+            actual_value = result.get(key)
+            if actual_value != expected_value:
+                print(f"  ❌ FAILED: {key}")
+                print(f"     Expected: {expected_value}")
+                print(f"     Got: {actual_value}")
+                test_passed = False
+                failed += 1
+                break
+
+        if test_passed:
+            print("  ✓ PASSED")
+            passed += 1
+        else:
+            # Show full result for debugging
+            print(f"     Full result: {json.dumps(result, indent=6)}")
+
+    print("\n" + "=" * 70)
+    print(f"Test Results: {passed} passed, {failed} failed out of {len(test_cases)} total")
+
+    if failed > 0:
+        print("\n❌ Some tests failed!")
+        sys.exit(1)
+    else:
+        print("\n✅ All tests passed!")
+        sys.exit(0)
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Analyze CVE exploitability for OpenClaw/NanoClaw deployments",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  # Analyze from JSON stdin
+  echo '{"cve_id":"CVE-2026-27488","cvss_score":7.3,"type":"ssrf"}' | python utils/analyze_exploitability.py --json
+
+  # Analyze with CVSS vector
+  echo '{"cve_id":"CVE-2026-1234","cvss_score":9.8,"cvss_vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H"}' \
+    | python utils/analyze_exploitability.py --json
+
+  # Run test cases
+  python utils/analyze_exploitability.py --test-cases
+
+  # Parse CVSS vector only
+  python utils/analyze_exploitability.py --parse-vector "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H"
+"""
+    )
+
+    parser.add_argument(
+        "--json",
+        action="store_true",
+        help="Read CVE data from stdin as JSON and output analysis"
+    )
+
+    parser.add_argument(
+        "--parse-vector",
+        type=str,
+        metavar="VECTOR",
+        help="Parse and display CVSS vector string"
+    )
+
+    parser.add_argument(
+        "--test-cases",
+        action="store_true",
+        help="Run built-in test cases to verify analyzer logic"
+    )
+
+    parser.add_argument(
+        "--check-exploits",
+        action="store_true",
+        help="Check references for publicly available exploits and adjust score accordingly"
+    )
+
+    args = parser.parse_args()
+
+    # Handle --parse-vector
+    if args.parse_vector:
+        metrics = parse_cvss_vector(args.parse_vector)
+        print(json.dumps(metrics, indent=2))
+        sys.exit(0)
+
+    # Handle --test-cases
+    if args.test_cases:
+        run_test_cases()
+        sys.exit(0)
+
+    # Handle --json (stdin)
+    if args.json:
+        try:
+            cve_data = json.load(sys.stdin)
+        except json.JSONDecodeError as e:
+            print(f"Error: Invalid JSON input: {e}", file=sys.stderr)
+            sys.exit(1)
+
+        result = analyze_exploitability(cve_data, check_exploits=args.check_exploits)
+        print(json.dumps(result, indent=2))
+        sys.exit(0)
+
+    # No action specified - show help
+    parser.print_help()
+    sys.exit(0)
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,420 @@
+# Exploitability Scoring Methodology
+
+## Overview
+
+ClawSec's exploitability scoring system provides context-aware vulnerability assessment specifically designed for AI agent deployments (OpenClaw/NanoClaw). Unlike generic CVSS scores that treat all environments equally, our scoring considers the unique attack surface and usage patterns of AI agents to reduce alert fatigue and prioritize actionable threats.
+
+## Scoring Levels
+
+| Level | Severity | Meaning |
+|---|---|---|
+| `high` | Critical/High | Exploitable in typical agent deployments, immediate attention required |
+| `medium` | Medium | May be exploitable depending on configuration, warrants investigation |
+| `low` | Low | Limited exploitability in agent context, low priority |
+| `unknown` | Unknown | Insufficient data to assess exploitability |
+
+## Scoring Factors
+
+### 1. CVSS Base Score (Baseline)
+
+The analysis starts with the CVSS base score as a foundation:
+
+- **CVSS ≥ 9.0**: Critical severity → initial score `high`
+- **CVSS 7.0-8.9**: High severity → initial score `high`
+- **CVSS 4.0-6.9**: Medium severity → initial score `medium`
+- **CVSS 1.0-3.9**: Low severity → initial score `low`
+- **No CVSS**: → initial score `unknown`
+
+### 2. Attack Vector Analysis (CVSS Metrics)
+
+The analyzer parses CVSS v2, v3.0, and v3.1 vectors to assess:
+
+#### Network Accessibility
+- **AV:N** (Network): Remotely exploitable over network
+- **AV:A** (Adjacent): Requires local network access
+- **AV:L** (Local): Requires local system access
+- **AV:P** (Physical): Requires physical access
+
+**Impact on agents**: Network-accessible vulnerabilities are elevated because agents typically run as network services or make external API calls.
+
+#### Authentication Requirements
+- **PR:N / Au:NONE**: No authentication required → elevates score
+- **PR:L / Au:SINGLE**: Low privileges required
+- **PR:H / Au:MULTIPLE**: High privileges required → reduces score
+
+**Impact on agents**: Unauthenticated exploits are critical for publicly exposed agent APIs.
+
+#### User Interaction
+- **UI:N**: No user interaction required → elevates score
+- **UI:R**: Requires user interaction → reduces score
+
+**Impact on agents**: Agents often operate autonomously, so vulnerabilities requiring user interaction are less critical.
+
+#### Attack Complexity
+- **AC:L**: Low complexity → elevates score
+- **AC:M / AC:H**: Medium/High complexity → neutral or reduces score
+
+**Impact on agents**: Low-complexity exploits are more likely to be automated and used in mass attacks.
+
+### 3. Vulnerability Type (Deployment Context)
+
+ClawSec adjusts scores based on how vulnerability types affect AI agent deployments:
+
+#### High-Risk Types in Agent Context
+
+**Remote Code Execution (RCE)**
+```
+Score: Always HIGH
+Rationale: RCE is critical in agent deployments
+```
+AI agents execute arbitrary code as part of their function. RCE vulnerabilities allow attackers to hijack agent execution flow, exfiltrate credentials, or pivot to other systems.
+
+**Server-Side Request Forgery (SSRF)**
+```
+Score: Elevated to HIGH if CVSS ≥ 6.0
+Rationale: SSRF affects agents making external requests
+```
+Agents frequently call external APIs, access internal services, and fetch remote resources. SSRF allows attackers to:
+- Access internal cloud metadata services (AWS IMDSv1, GCP metadata)
+- Pivot to internal networks
+- Exfiltrate data through DNS tunneling
+
+**Path Traversal / Directory Traversal**
+```
+Score: Elevated to HIGH if CVSS ≥ 6.0
+Rationale: Path traversal affects agents with file access
+```
+Agents read files, execute scripts, and manage codebases. Path traversal enables:
+- Reading sensitive configuration files (.env, credentials)
+- Accessing SSH keys, API tokens
+- Overwriting critical system files
+
+**Command Injection**
+```
+Score: Always HIGH
+Rationale: Command injection is critical in agent deployments
+```
+Similar to RCE, agents often execute shell commands to interact with systems. Command injection allows full system compromise.
+
+#### Medium-Risk Types
+
+**Prototype Pollution (Node.js)**
+```
+Score: Elevated from LOW to MEDIUM
+Rationale: Prototype pollution can escalate in Node.js agents
+```
+Many agent frameworks run on Node.js. Prototype pollution can lead to:
+- Bypass of authentication checks
+- Privilege escalation
+- Denial of service
+
+**SQL Injection / NoSQL Injection**
+```
+Score: Elevated to HIGH if network-accessible and unauthenticated
+Rationale: Injection affects agents with database access
+```
+Agents that store conversation history, user data, or tool results in databases are vulnerable to injection attacks.
+
+#### Lower-Risk Types
+
+**Cross-Site Scripting (XSS)**
+```
+Score: Reduced to MEDIUM if not network-accessible
+Rationale: XSS has limited impact in headless agents
+```
+Agents typically don't render HTML in browsers, reducing XSS impact. However, XSS in agent management UIs or chat interfaces remains a concern.
+
+### 4. Exploit Availability
+
+When `--check-exploits` is enabled, the analyzer checks reference URLs for public exploits:
+
+**Exploit Indicators:**
+- exploit-db.com / exploit-database.com
+- packetstormsecurity.com
+- github.com/exploit, github.com/poc
+- metasploit framework modules
+- URLs containing "/exploit", "/poc", "/proof-of-concept"
+
+**Score Elevation:**
+- `low` → `medium` (exploit available)
+- `medium` → `high` (exploit available)
+- `unknown` → `medium` (exploit available + CVSS > 0)
+
+**Rationale**: Public exploits lower the skill barrier for attackers and increase the likelihood of automated exploitation.
+
+## Scoring Algorithm
+
+The analyzer follows this decision tree:
+
+```
+1. Parse CVSS score → set baseline (high/medium/low/unknown)
+2. Parse CVSS vector → analyze attack characteristics
+3. Adjust for attack vector:
+   - Network-accessible + no auth + no UI → elevate to HIGH
+   - Local-only access → reduce HIGH to MEDIUM
+4. Adjust for vulnerability type:
+   - Check against agent-specific risk categories
+   - Elevate or reduce score based on deployment context
+5. Check for public exploits (if enabled):
+   - Elevate score if exploits detected
+6. Generate rationale explaining the final score
+```
+
+## Examples
+
+### Example 1: Critical RCE (High Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-12345",
+  "cvss_score": 9.8,
+  "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
+  "type": "remote_code_execution",
+  "description": "Unauthenticated RCE in Express.js framework"
+}
+```
+
+**Analysis Output:**
+```json
+{
+  "exploitability_score": "high",
+  "exploitability_rationale": "Critical CVSS score (9.8); remotely exploitable without authentication; RCE is critical in agent deployments"
+}
+```
+
+**Why HIGH**: Critical CVSS + network accessible + no auth + RCE type.
+
+### Example 2: SSRF in Agent API (High Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-23456",
+  "cvss_score": 7.3,
+  "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L",
+  "type": "server_side_request_forgery",
+  "description": "SSRF in webhook handler allows internal network access"
+}
+```
+
+**Analysis Output:**
+```json
+{
+  "exploitability_score": "high",
+  "exploitability_rationale": "High CVSS score (7.3); remotely exploitable without authentication; SSRF affects agents making external requests"
+}
+```
+
+**Why HIGH**: SSRF is critical for agents that make API calls (most do). Network-accessible without authentication elevates risk.
+
+### Example 3: Path Traversal with Public Exploit (High Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-34567",
+  "cvss_score": 6.5,
+  "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N",
+  "type": "path_traversal",
+  "references": [
+    "https://exploit-db.com/exploits/51234",
+    "https://nvd.nist.gov/vuln/detail/CVE-2024-34567"
+  ]
+}
+```
+
+**Analysis Output (with --check-exploits):**
+```json
+{
+  "exploitability_score": "high",
+  "exploitability_rationale": "Medium CVSS score (6.5); network accessible; path traversal affects agents with file access; public exploit available (1 source)"
+}
+```
+
+**Why HIGH**: Path traversal + agent file access + public exploit elevates medium CVSS to high exploitability.
+
+### Example 4: XSS in Agent UI (Medium Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-45678",
+  "cvss_score": 7.1,
+  "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:L",
+  "type": "cross_site_scripting",
+  "description": "Stored XSS in agent management dashboard"
+}
+```
+
+**Analysis Output:**
+```json
+{
+  "exploitability_score": "medium",
+  "exploitability_rationale": "High CVSS score (7.1); network accessible; XSS has limited impact in headless agents"
+}
+```
+
+**Why MEDIUM**: Despite high CVSS, XSS is less critical in agent deployments (headless operation). Requires user interaction.
+
+### Example 5: Local Privilege Escalation (Medium Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-56789",
+  "cvss_score": 8.8,
+  "cvss_vector": "CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H",
+  "type": "privilege_escalation",
+  "description": "Local privilege escalation via symbolic link attack"
+}
+```
+
+**Analysis Output:**
+```json
+{
+  "exploitability_score": "medium",
+  "exploitability_rationale": "High CVSS score (8.8); requires local access"
+}
+```
+
+**Why MEDIUM**: Despite high CVSS, requires local access. Agents typically run in containerized/sandboxed environments where local escalation has limited impact.
+
+### Example 6: Prototype Pollution with Exploit (High Exploitability)
+
+```json
+{
+  "cve_id": "CVE-2024-67890",
+  "cvss_score": 5.3,
+  "cvss_vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:N",
+  "type": "prototype_pollution",
+  "description": "Prototype pollution in lodash merge function",
+  "references": [
+    "https://github.com/exploit/prototype-pollution-poc",
+    "https://snyk.io/vuln/SNYK-JS-LODASH-1234567"
+  ]
+}
+```
+
+**Analysis Output (with --check-exploits):**
+```json
+{
+  "exploitability_score": "high",
+  "exploitability_rationale": "Medium CVSS score (5.3); remotely exploitable without authentication; prototype pollution can escalate in Node.js agents; public exploit available (1 source)"
+}
+```
+
+**Why HIGH**: Prototype pollution in Node.js agents + public exploit + network-accessible without auth = high risk despite moderate CVSS.
+
+## Usage in ClawSec Workflows
+
+### Automated Scoring (NVD Feed)
+
+The `poll-nvd-cves.yml` workflow automatically scores new CVEs:
+
+```bash
+# Workflow step
+python utils/analyze_exploitability.py --json --check-exploits < cve-data.json
+```
+
+Advisories in `advisories/feed.json` can include:
+```json
+{
+  "id": "CVE-2024-12345",
+  "severity": "high",
+  "exploitability_score": "high",
+  "exploitability_rationale": "Critical CVSS score (9.8); remotely exploitable without authentication; RCE is critical in agent deployments",
+  "attack_vector_analysis": {
+    "is_network_accessible": true,
+    "requires_authentication": false,
+    "requires_user_interaction": false,
+    "complexity": "low"
+  }
+}
+```
+
+### Manual Analysis
+
+Security researchers can analyze CVEs manually:
+
+```bash
+# Basic analysis
+echo '{"cve_id":"CVE-2024-12345","cvss_score":7.3,"type":"ssrf"}' | \
+  python utils/analyze_exploitability.py --json
+
+# With exploit detection
+echo '{"cve_id":"CVE-2024-12345","cvss_score":7.3,"references":["https://exploit-db.com/exploits/51234"]}' | \
+  python utils/analyze_exploitability.py --json --check-exploits
+```
+
+### Filtering by Exploitability
+
+Users can filter advisories by exploitability score:
+
+```bash
+# Get only high-exploitability advisories
+curl -s https://clawsec.prompt.security/feed.json | \
+  jq '.advisories[] | select(.exploitability_score == "high")'
+
+# Prioritize by exploitability and severity
+curl -s https://clawsec.prompt.security/feed.json | \
+  jq '[.advisories[] | select(.exploitability_score == "high" and .severity == "critical")] | sort_by(.cvss_score) | reverse'
+```
+
+## Backfilling Existing Advisories (Historical Maintenance)
+
+`scripts/backfill-exploitability.sh` is retained as a historical maintainer utility for one-off repository maintenance.
+It is not the primary path for normal advisory generation.
+
+Preferred paths:
+
+1. CI canonical path: run the NVD workflow with init/reset to rebuild advisories from NVD and sign artifacts in pipeline.
+2. Local developer path: run `./scripts/populate-local-feed.sh --force` to repopulate local feeds with exploitability context.
+
+Use backfill only when explicitly repairing legacy feed content that already exists in-repo.
+
+## Community Contributions
+
+Community members can submit exploitability assessments:
+
+1. **Report via GitHub Issue**: Use the advisory template to report CVEs with exploitability context
+2. **Automated Analysis**: The `community-advisory.yml` workflow automatically scores community-reported CVEs
+3. **Manual Review**: Maintainers review and approve exploitability assessments
+4. **Feed Update**: Approved advisories are added to the feed with exploitability scores
+
+## Limitations and Future Work
+
+### Current Limitations
+
+1. **Static Analysis**: Scoring is based on CVE metadata, not dynamic runtime analysis
+2. **No Version Detection**: Doesn't check if specific versions are vulnerable
+3. **Binary Classification**: Doesn't consider partial mitigations or defense-in-depth
+4. **Limited Context**: Doesn't know exact agent configuration or deployed tools
+
+### Future Enhancements
+
+1. **EPSS Integration**: Incorporate EPSS (Exploit Prediction Scoring System) probability scores
+2. **KEV Matching**: Cross-reference with CISA KEV (Known Exploited Vulnerabilities) catalog
+3. **Agent Profiling**: Consider deployed agent capabilities and exposed APIs
+4. **Mitigation Detection**: Check for WAF rules, sandboxing, or other compensating controls
+5. **ML-Based Scoring**: Use machine learning to predict exploitability based on historical data
+
+## References
+
+- **CVSS v3.1 Specification**: [https://www.first.org/cvss/v3.1/specification-document](https://www.first.org/cvss/v3.1/specification-document)
+- **CVSS v2 Guide**: [https://www.first.org/cvss/v2/guide](https://www.first.org/cvss/v2/guide)
+- **EPSS**: [https://www.first.org/epss/](https://www.first.org/epss/)
+- **CISA KEV**: [https://www.cisa.gov/known-exploited-vulnerabilities-catalog](https://www.cisa.gov/known-exploited-vulnerabilities-catalog)
+- **NVD API**: [https://nvd.nist.gov/developers/vulnerabilities](https://nvd.nist.gov/developers/vulnerabilities)
+
+## Contributing
+
+To improve the exploitability scoring methodology:
+
+1. **Submit Test Cases**: Add test cases to `utils/analyze_exploitability.py`
+2. **Report False Positives/Negatives**: Open GitHub issues with CVE examples
+3. **Propose Scoring Adjustments**: Submit PRs with rationale and examples
+4. **Share Agent Context**: Contribute agent-specific vulnerability patterns
+
+See [CONTRIBUTING.md](../CONTRIBUTING.md) for detailed contribution guidelines.
+
+---
+
+**Maintained by**: [Prompt Security](https://prompt.security)
+**License**: AGPL-3.0-or-later
+**Last Updated**: 2026-03-01