[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6NTP0TJTNRCK5WGRV5CWFXQ":3,"skill-same:01M6NTP0TJTNRCK5WGRV5CWFXQ":75},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":43,"linkOnly":44,"markdownContent":45,"markdownBody":46,"frontMatter":47,"refName":33,"commitSha":56,"skillDirectory":57,"forksCount":13,"repositoryHtmlUrl":58,"repositoryHomepageUrl":6,"ownerHtmlUrl":59,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":60,"install":61,"resources":74},"01M6NTP0TJTNRCK5WGRV5CWFXQ","ai-security-hardening",null,"Harden AI\u002FLLM deployments against prompt injection, data exfiltration, model theft, and supply chain attacks.","duc-duy-1612\u002FOnline-Delivery-Systems-Web","duc-duy-1612","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F192231384?v=4","01M6NSDKQQDKVTHZ8DZMZ1VYY6","01M6NSDKW8DDVTQHBT1HBDDJPT",0,"MIT","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fai-security-hardening\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42],"ai security","security","hardening","security hardening","ai security hardening","relevant","scripts","scope","chain","active","main","community","only","model","devops-security","data","exfiltration","assessment","safe","prompt injection","2026-11-10 10:12:30",false,"---\nname: ai-security-hardening\ndescription: Harden AI\u002FLLM deployments against prompt injection, data exfiltration,\n  model theft, and supply chain attacks.\ncategory: security\nrisk: safe\nsource: https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills\nsource_repo: BagelHole\u002FDevOps-Security-Agent-Skills\nsource_type: community\ndate_added: '2026-09-20'\nlicense: MIT\nlicense_source: https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills\u002Fblob\u002Fmain\u002FLICENSE\ncompatibility: Requires the relevant security tooling (scanners, vault CLIs) and an\n  authorized scope for any active assessment. Docs-only; helper scripts and templates\n  not bundled.\nmetadata:\n  author: devops-skills\n  version: '1.0'\n---\n\n# AI Security Hardening\n\nSecure LLM and AI systems against prompt injection, jailbreaks, data leakage, and supply chain threats in production environments.\n\n## When to Use This Skill\n\nUse this skill when:\n- Deploying an LLM-powered application handling sensitive user data\n- Protecting against prompt injection attacks in AI agents\n- Implementing output filtering and content moderation\n- Securing model weights and API endpoints from theft\n- Achieving SOC2 or ISO 27001 compliance for AI systems\n\n## AI-Specific Threat Model\n\n```\nThreat                    Risk                          Control\n─────────────────────────────────────────────────────────────────────\nPrompt injection          System prompt override         Input sanitization, separate context\nData exfiltration         PII in model outputs           Output filtering, DLP scanning\nJailbreaking             Policy bypass                  Content moderation, guardrails\nModel theft               Weight extraction via API      Rate limiting, access controls\nTraining data poisoning   Backdoored fine-tuned model    Dataset validation, provenance\nSupply chain attack       Malicious model weights        Signature verification, scanning\nInsecure output           XSS\u002FSQLi from LLM response     Output encoding, parameterized queries\n```\n\n## Prompt Injection Defense\n\n```python\nimport re\nfrom typing import Optional\n\nINJECTION_PATTERNS = [\n    r\"ignore\\s+(all\\s+)?(previous|prior|above)\\s+instructions\",\n    r\"you\\s+are\\s+now\\s+\",\n    r\"new\\s+instructions?:\",\n    r\"system\\s+prompt\",\n    r\"forget\\s+everything\",\n    r\"act\\s+as\\s+\",\n    r\"jailbreak\",\n    r\"dan\\s+mode\",\n    r\"\u003C\\s*system\\s*>\",\n    r\"\\[INST\\]\",\n]\n\ndef detect_prompt_injection(user_input: str) -> tuple[bool, Optional[str]]:\n    \"\"\"Return (is_suspicious, matched_pattern).\"\"\"\n    normalized = user_input.lower().strip()\n    for pattern in INJECTION_PATTERNS:\n        if re.search(pattern, normalized, re.IGNORECASE):\n            return True, pattern\n    return False, None\n\ndef sanitize_user_input(user_input: str, max_length: int = 4000) -> str:\n    \"\"\"Sanitize input before passing to LLM.\"\"\"\n    # Truncate\n    user_input = user_input[:max_length]\n\n    # Remove null bytes and control characters\n    user_input = re.sub(r'[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]', '', user_input)\n\n    # Check for injection\n    suspicious, pattern = detect_prompt_injection(user_input)\n    if suspicious:\n        raise ValueError(f\"Potential prompt injection detected: {pattern}\")\n\n    return user_input\n```\n\n## Guardrails with NeMo Guardrails\n\n```python\n# guardrails.yaml\nfrom nemoguardrails import RailsConfig, LLMRails\n\nconfig = RailsConfig.from_path(\".\u002Fguardrails-config\")\nrails = LLMRails(config)\n\nasync def safe_llm_call(user_message: str) -> str:\n    response = await rails.generate_async(\n        messages=[{\"role\": \"user\", \"content\": user_message}]\n    )\n    return response[\"content\"]\n```\n\n```yaml\n# guardrails-config\u002Fconfig.yml\nmodels:\n  - type: main\n    engine: openai\n    model: gpt-4o-mini\n\nrails:\n  input:\n    flows:\n      - check jailbreak\n      - check sensitive data\n  output:\n    flows:\n      - check output for PII\n      - check output for harmful content\n```\n\n## Output Filtering & PII Scrubbing\n\n```python\nimport re\nfrom presidio_analyzer import AnalyzerEngine\nfrom presidio_anonymizer import AnonymizerEngine\n\nanalyzer = AnalyzerEngine()\nanonymizer = AnonymizerEngine()\n\nPII_ENTITIES = [\"PERSON\", \"EMAIL_ADDRESS\", \"PHONE_NUMBER\", \"CREDIT_CARD\",\n                \"US_SSN\", \"IBAN_CODE\", \"IP_ADDRESS\", \"LOCATION\"]\n\ndef scrub_pii_from_output(text: str) -> str:\n    \"\"\"Remove PII from LLM output before returning to user.\"\"\"\n    results = analyzer.analyze(text=text, entities=PII_ENTITIES, language=\"en\")\n    if not results:\n        return text\n    anonymized = anonymizer.anonymize(text=text, analyzer_results=results)\n    return anonymized.text\n\ndef validate_output_safety(output: str) -> bool:\n    \"\"\"Check output doesn't contain prompt injection artifacts.\"\"\"\n    dangerous_patterns = [\n        r\"\u003C\\s*script\\s*>\",         # XSS\n        r\"javascript:\",             # XSS\n        r\";\\s*(DROP|DELETE|INSERT)\",# SQLi\n        r\"\\$\\{.*\\}\",               # template injection\n        r\"`.*`\",                   # command injection in some contexts\n    ]\n    for pattern in dangerous_patterns:\n        if re.search(pattern, output, re.IGNORECASE):\n            return False\n    return True\n```\n\n## API Security for LLM Endpoints\n\n```python\nfrom fastapi import FastAPI, HTTPException, Depends, Request\nfrom fastapi.security import HTTPBearer, HTTPAuthorizationCredentials\nimport jwt\nimport time\nfrom collections import defaultdict\n\napp = FastAPI()\nsecurity = HTTPBearer()\n\n# Rate limiting (per API key)\nrequest_counts = defaultdict(list)\n\ndef rate_limit(api_key: str, max_requests: int = 100, window_seconds: int = 60):\n    now = time.time()\n    requests = request_counts[api_key]\n    # Remove old requests outside window\n    request_counts[api_key] = [t for t in requests if now - t \u003C window_seconds]\n    if len(request_counts[api_key]) >= max_requests:\n        raise HTTPException(status_code=429, detail=\"Rate limit exceeded\")\n    request_counts[api_key].append(now)\n\nasync def verify_token(\n    credentials: HTTPAuthorizationCredentials = Depends(security)\n) -> dict:\n    try:\n        payload = jwt.decode(credentials.credentials, SECRET_KEY, algorithms=[\"HS256\"])\n        rate_limit(payload[\"sub\"])\n        return payload\n    except jwt.ExpiredSignatureError:\n        raise HTTPException(status_code=401, detail=\"Token expired\")\n    except jwt.InvalidTokenError:\n        raise HTTPException(status_code=401, detail=\"Invalid token\")\n\n@app.post(\"\u002Fv1\u002Fchat\u002Fcompletions\")\nasync def chat(request: Request, token: dict = Depends(verify_token)):\n    body = await request.json()\n\n    # Input validation\n    user_msg = body.get(\"messages\", [{}])[-1].get(\"content\", \"\")\n    try:\n        safe_input = sanitize_user_input(user_msg)\n    except ValueError as e:\n        raise HTTPException(status_code=400, detail=str(e))\n\n    # Call LLM and scrub output\n    response = await call_llm(safe_input, token[\"scope\"])\n    response[\"choices\"][0][\"message\"][\"content\"] = scrub_pii_from_output(\n        response[\"choices\"][0][\"message\"][\"content\"]\n    )\n    return response\n```\n\n## Model Weight Security\n\n```bash\n# Verify model weights with SHA-256 hash before loading\nMODEL_DIR=\".\u002Fmodels\u002Fllama-3.1-8b\"\nEXPECTED_HASH=\"sha256:abc123...\"\n\n# Generate hash of downloaded model\nactual_hash=$(find \"$MODEL_DIR\" -name \"*.safetensors\" | sort | xargs sha256sum | sha256sum)\necho \"Model hash: $actual_hash\"\n\n# Compare (automate in CI\u002FCD)\nif [ \"$actual_hash\" != \"$EXPECTED_HASH\" ]; then\n  echo \"ERROR: Model hash mismatch — possible tampering!\"\n  exit 1\nfi\n\n# Scan model files for embedded malware (ModelScan)\npip install modelscan\nmodelscan scan -p \"$MODEL_DIR\"\n```\n\n## Network Isolation for AI Services\n\n```yaml\n# Kubernetes NetworkPolicy — isolate LLM API\napiVersion: networking.k8s.io\u002Fv1\nkind: NetworkPolicy\nmetadata:\n  name: llm-api-isolation\n  namespace: ai-services\nspec:\n  podSelector:\n    matchLabels:\n      app: vllm\n  policyTypes:\n  - Ingress\n  - Egress\n  ingress:\n  - from:\n    - namespaceSelector:\n        matchLabels:\n          name: backend           # only backend can call LLM\n    ports:\n    - protocol: TCP\n      port: 8000\n  egress:\n  - to:\n    - namespaceSelector:\n        matchLabels:\n          name: monitoring        # metrics only\n    ports:\n    - protocol: TCP\n      port: 9090\n  # Block egress to internet — prevent data exfiltration\n  # (allow only internal cluster traffic)\n```\n\n## Audit Logging\n\n```python\nimport structlog\nfrom datetime import datetime, timezone\n\naudit_log = structlog.get_logger(\"ai.audit\")\n\ndef log_llm_interaction(\n    user_id: str,\n    session_id: str,\n    model: str,\n    prompt_tokens: int,\n    completion_tokens: int,\n    was_filtered: bool,\n    injection_detected: bool,\n):\n    audit_log.info(\n        \"llm_interaction\",\n        timestamp=datetime.now(timezone.utc).isoformat(),\n        user_id=user_id,\n        session_id=session_id,\n        model=model,\n        prompt_tokens=prompt_tokens,\n        completion_tokens=completion_tokens,\n        was_filtered=was_filtered,\n        injection_detected=injection_detected,\n        # DO NOT log prompt\u002Fcompletion content — PII risk\n    )\n```\n\n## Common Issues\n\n| Issue | Cause | Fix |\n|-------|-------|-----|\n| False positive injection blocks | Overly broad regex | Tune patterns; use ML-based classifier for high-traffic |\n| PII in model outputs | Model trained on PII data | Add Presidio scrubbing to output layer |\n| API key leakage | Keys in logs or responses | Mask keys in logging; use vault for key storage |\n| Model weight tampering | Unverified downloads | Always verify SHA-256; use `modelscan` |\n| Rate limit bypass | Per-IP not per-user | Rate limit on authenticated user ID, not IP |\n\n## Best Practices\n\n- Never log raw prompts or completions — they may contain PII or sensitive data.\n- Treat LLM output as untrusted input — always encode before rendering in HTML.\n- Use network policies to prevent LLM pods from making outbound internet calls.\n- Rotate API keys quarterly; use short-lived JWT tokens for service-to-service auth.\n- Run `modelscan` on any model downloaded from the internet before serving.\n\n## Related Skills\n\n- hashicorp-vault (`hashicorp-vault`) - Secrets management for API keys\n- [network-security] (..\u002F..\u002Fnetwork\u002F) - Network-level controls\n- linux-hardening (`linux-hardening`) - Host hardening\n- agent-observability (`agent-observability`) - AI audit logging\n- llm-gateway (`llm-gateway`) - Centralized access control\n\n## Limitations\n\n- Apply guidance only within authorized scope; test destructive steps in non-production first.\n- Docs-only import: upstream scripts and templates not bundled.\n\n### Example\n\n```bash\n# Read-only first: inventory before any active step.\nwhich \u003Ctool> && \u003Ctool> --help | head -n 20\n```\n\n> Adapted from [BagelHole\u002FDevOps-Security-Agent-Skills](https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills) (MIT); frontmatter, When to Use\u002FLimitations, and safety boundaries added for upstream compliance. Docs-only import: helper scripts and templates not bundled.\n","# AI Security Hardening\n\nSecure LLM and AI systems against prompt injection, jailbreaks, data leakage, and supply chain threats in production environments.\n\n## When to Use This Skill\n\nUse this skill when:\n- Deploying an LLM-powered application handling sensitive user data\n- Protecting against prompt injection attacks in AI agents\n- Implementing output filtering and content moderation\n- Securing model weights and API endpoints from theft\n- Achieving SOC2 or ISO 27001 compliance for AI systems\n\n## AI-Specific Threat Model\n\n```\nThreat                    Risk                          Control\n─────────────────────────────────────────────────────────────────────\nPrompt injection          System prompt override         Input sanitization, separate context\nData exfiltration         PII in model outputs           Output filtering, DLP scanning\nJailbreaking             Policy bypass                  Content moderation, guardrails\nModel theft               Weight extraction via API      Rate limiting, access controls\nTraining data poisoning   Backdoored fine-tuned model    Dataset validation, provenance\nSupply chain attack       Malicious model weights        Signature verification, scanning\nInsecure output           XSS\u002FSQLi from LLM response     Output encoding, parameterized queries\n```\n\n## Prompt Injection Defense\n\n```python\nimport re\nfrom typing import Optional\n\nINJECTION_PATTERNS = [\n    r\"ignore\\s+(all\\s+)?(previous|prior|above)\\s+instructions\",\n    r\"you\\s+are\\s+now\\s+\",\n    r\"new\\s+instructions?:\",\n    r\"system\\s+prompt\",\n    r\"forget\\s+everything\",\n    r\"act\\s+as\\s+\",\n    r\"jailbreak\",\n    r\"dan\\s+mode\",\n    r\"\u003C\\s*system\\s*>\",\n    r\"\\[INST\\]\",\n]\n\ndef detect_prompt_injection(user_input: str) -> tuple[bool, Optional[str]]:\n    \"\"\"Return (is_suspicious, matched_pattern).\"\"\"\n    normalized = user_input.lower().strip()\n    for pattern in INJECTION_PATTERNS:\n        if re.search(pattern, normalized, re.IGNORECASE):\n            return True, pattern\n    return False, None\n\ndef sanitize_user_input(user_input: str, max_length: int = 4000) -> str:\n    \"\"\"Sanitize input before passing to LLM.\"\"\"\n    # Truncate\n    user_input = user_input[:max_length]\n\n    # Remove null bytes and control characters\n    user_input = re.sub(r'[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]', '', user_input)\n\n    # Check for injection\n    suspicious, pattern = detect_prompt_injection(user_input)\n    if suspicious:\n        raise ValueError(f\"Potential prompt injection detected: {pattern}\")\n\n    return user_input\n```\n\n## Guardrails with NeMo Guardrails\n\n```python\n# guardrails.yaml\nfrom nemoguardrails import RailsConfig, LLMRails\n\nconfig = RailsConfig.from_path(\".\u002Fguardrails-config\")\nrails = LLMRails(config)\n\nasync def safe_llm_call(user_message: str) -> str:\n    response = await rails.generate_async(\n        messages=[{\"role\": \"user\", \"content\": user_message}]\n    )\n    return response[\"content\"]\n```\n\n```yaml\n# guardrails-config\u002Fconfig.yml\nmodels:\n  - type: main\n    engine: openai\n    model: gpt-4o-mini\n\nrails:\n  input:\n    flows:\n      - check jailbreak\n      - check sensitive data\n  output:\n    flows:\n      - check output for PII\n      - check output for harmful content\n```\n\n## Output Filtering & PII Scrubbing\n\n```python\nimport re\nfrom presidio_analyzer import AnalyzerEngine\nfrom presidio_anonymizer import AnonymizerEngine\n\nanalyzer = AnalyzerEngine()\nanonymizer = AnonymizerEngine()\n\nPII_ENTITIES = [\"PERSON\", \"EMAIL_ADDRESS\", \"PHONE_NUMBER\", \"CREDIT_CARD\",\n                \"US_SSN\", \"IBAN_CODE\", \"IP_ADDRESS\", \"LOCATION\"]\n\ndef scrub_pii_from_output(text: str) -> str:\n    \"\"\"Remove PII from LLM output before returning to user.\"\"\"\n    results = analyzer.analyze(text=text, entities=PII_ENTITIES, language=\"en\")\n    if not results:\n        return text\n    anonymized = anonymizer.anonymize(text=text, analyzer_results=results)\n    return anonymized.text\n\ndef validate_output_safety(output: str) -> bool:\n    \"\"\"Check output doesn't contain prompt injection artifacts.\"\"\"\n    dangerous_patterns = [\n        r\"\u003C\\s*script\\s*>\",         # XSS\n        r\"javascript:\",             # XSS\n        r\";\\s*(DROP|DELETE|INSERT)\",# SQLi\n        r\"\\$\\{.*\\}\",               # template injection\n        r\"`.*`\",                   # command injection in some contexts\n    ]\n    for pattern in dangerous_patterns:\n        if re.search(pattern, output, re.IGNORECASE):\n            return False\n    return True\n```\n\n## API Security for LLM Endpoints\n\n```python\nfrom fastapi import FastAPI, HTTPException, Depends, Request\nfrom fastapi.security import HTTPBearer, HTTPAuthorizationCredentials\nimport jwt\nimport time\nfrom collections import defaultdict\n\napp = FastAPI()\nsecurity = HTTPBearer()\n\n# Rate limiting (per API key)\nrequest_counts = defaultdict(list)\n\ndef rate_limit(api_key: str, max_requests: int = 100, window_seconds: int = 60):\n    now = time.time()\n    requests = request_counts[api_key]\n    # Remove old requests outside window\n    request_counts[api_key] = [t for t in requests if now - t \u003C window_seconds]\n    if len(request_counts[api_key]) >= max_requests:\n        raise HTTPException(status_code=429, detail=\"Rate limit exceeded\")\n    request_counts[api_key].append(now)\n\nasync def verify_token(\n    credentials: HTTPAuthorizationCredentials = Depends(security)\n) -> dict:\n    try:\n        payload = jwt.decode(credentials.credentials, SECRET_KEY, algorithms=[\"HS256\"])\n        rate_limit(payload[\"sub\"])\n        return payload\n    except jwt.ExpiredSignatureError:\n        raise HTTPException(status_code=401, detail=\"Token expired\")\n    except jwt.InvalidTokenError:\n        raise HTTPException(status_code=401, detail=\"Invalid token\")\n\n@app.post(\"\u002Fv1\u002Fchat\u002Fcompletions\")\nasync def chat(request: Request, token: dict = Depends(verify_token)):\n    body = await request.json()\n\n    # Input validation\n    user_msg = body.get(\"messages\", [{}])[-1].get(\"content\", \"\")\n    try:\n        safe_input = sanitize_user_input(user_msg)\n    except ValueError as e:\n        raise HTTPException(status_code=400, detail=str(e))\n\n    # Call LLM and scrub output\n    response = await call_llm(safe_input, token[\"scope\"])\n    response[\"choices\"][0][\"message\"][\"content\"] = scrub_pii_from_output(\n        response[\"choices\"][0][\"message\"][\"content\"]\n    )\n    return response\n```\n\n## Model Weight Security\n\n```bash\n# Verify model weights with SHA-256 hash before loading\nMODEL_DIR=\".\u002Fmodels\u002Fllama-3.1-8b\"\nEXPECTED_HASH=\"sha256:abc123...\"\n\n# Generate hash of downloaded model\nactual_hash=$(find \"$MODEL_DIR\" -name \"*.safetensors\" | sort | xargs sha256sum | sha256sum)\necho \"Model hash: $actual_hash\"\n\n# Compare (automate in CI\u002FCD)\nif [ \"$actual_hash\" != \"$EXPECTED_HASH\" ]; then\n  echo \"ERROR: Model hash mismatch — possible tampering!\"\n  exit 1\nfi\n\n# Scan model files for embedded malware (ModelScan)\npip install modelscan\nmodelscan scan -p \"$MODEL_DIR\"\n```\n\n## Network Isolation for AI Services\n\n```yaml\n# Kubernetes NetworkPolicy — isolate LLM API\napiVersion: networking.k8s.io\u002Fv1\nkind: NetworkPolicy\nmetadata:\n  name: llm-api-isolation\n  namespace: ai-services\nspec:\n  podSelector:\n    matchLabels:\n      app: vllm\n  policyTypes:\n  - Ingress\n  - Egress\n  ingress:\n  - from:\n    - namespaceSelector:\n        matchLabels:\n          name: backend           # only backend can call LLM\n    ports:\n    - protocol: TCP\n      port: 8000\n  egress:\n  - to:\n    - namespaceSelector:\n        matchLabels:\n          name: monitoring        # metrics only\n    ports:\n    - protocol: TCP\n      port: 9090\n  # Block egress to internet — prevent data exfiltration\n  # (allow only internal cluster traffic)\n```\n\n## Audit Logging\n\n```python\nimport structlog\nfrom datetime import datetime, timezone\n\naudit_log = structlog.get_logger(\"ai.audit\")\n\ndef log_llm_interaction(\n    user_id: str,\n    session_id: str,\n    model: str,\n    prompt_tokens: int,\n    completion_tokens: int,\n    was_filtered: bool,\n    injection_detected: bool,\n):\n    audit_log.info(\n        \"llm_interaction\",\n        timestamp=datetime.now(timezone.utc).isoformat(),\n        user_id=user_id,\n        session_id=session_id,\n        model=model,\n        prompt_tokens=prompt_tokens,\n        completion_tokens=completion_tokens,\n        was_filtered=was_filtered,\n        injection_detected=injection_detected,\n        # DO NOT log prompt\u002Fcompletion content — PII risk\n    )\n```\n\n## Common Issues\n\n| Issue | Cause | Fix |\n|-------|-------|-----|\n| False positive injection blocks | Overly broad regex | Tune patterns; use ML-based classifier for high-traffic |\n| PII in model outputs | Model trained on PII data | Add Presidio scrubbing to output layer |\n| API key leakage | Keys in logs or responses | Mask keys in logging; use vault for key storage |\n| Model weight tampering | Unverified downloads | Always verify SHA-256; use `modelscan` |\n| Rate limit bypass | Per-IP not per-user | Rate limit on authenticated user ID, not IP |\n\n## Best Practices\n\n- Never log raw prompts or completions — they may contain PII or sensitive data.\n- Treat LLM output as untrusted input — always encode before rendering in HTML.\n- Use network policies to prevent LLM pods from making outbound internet calls.\n- Rotate API keys quarterly; use short-lived JWT tokens for service-to-service auth.\n- Run `modelscan` on any model downloaded from the internet before serving.\n\n## Related Skills\n\n- hashicorp-vault (`hashicorp-vault`) - Secrets management for API keys\n- [network-security] (..\u002F..\u002Fnetwork\u002F) - Network-level controls\n- linux-hardening (`linux-hardening`) - Host hardening\n- agent-observability (`agent-observability`) - AI audit logging\n- llm-gateway (`llm-gateway`) - Centralized access control\n\n## Limitations\n\n- Apply guidance only within authorized scope; test destructive steps in non-production first.\n- Docs-only import: upstream scripts and templates not bundled.\n\n### Example\n\n```bash\n# Read-only first: inventory before any active step.\nwhich \u003Ctool> && \u003Ctool> --help | head -n 20\n```\n\n> Adapted from [BagelHole\u002FDevOps-Security-Agent-Skills](https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills) (MIT); frontmatter, When to Use\u002FLimitations, and safety boundaries added for upstream compliance. Docs-only import: helper scripts and templates not bundled.\n",{"name":5,"risk":41,"source":48,"license":14,"category":24,"metadata":49,"date_added":52,"description":7,"source_repo":53,"source_type":34,"compatibility":54,"license_source":55},"https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills",{"author":50,"version":51},"devops-skills","1.0","2026-09-20","BagelHole\u002FDevOps-Security-Agent-Skills","Requires the relevant security tooling (scanners, vault CLIs) and an authorized scope for any active assessment. Docs-only; helper scripts and templates not bundled.","https:\u002F\u002Fgithub.com\u002FBagelHole\u002FDevOps-Security-Agent-Skills\u002Fblob\u002Fmain\u002FLICENSE","f3312ba749561b6159f477dbfc9313c875d40a56",".agents\u002Fskills_archive\u002Fai-security-hardening","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612","2026-10-25 23:50:58",{"source":8,"skillName":5,"commands":62,"prompt":72,"zipAvailable":73},[63,66,69],{"tool":64,"command":65},"npx","npx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill ai-security-hardening",{"tool":67,"command":68},"bunx","bunx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill ai-security-hardening",{"tool":70,"command":71},"pnpm","pnpm dlx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill ai-security-hardening","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fai-security-hardening\u002FSKILL.md",true,[],[76,96,119,143,171,195],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":85,"publishedAt":43},"01M6Q0FA5FZ410Z33Z3TFQ4DBA","zero-trust","Implement zero-trust network architecture. Configure identity-based access, micro-segmentation, and continuous verification. Use when implementing modern security architectures.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fzero-trust\u002FSKILL.md",[82],{"categoryNo":83,"categoryName":84,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000011","运维与部署",[86,87,88,89,40,90,78,91,37,92,93,35,34,33,32,94,30,29,28,95],"ZeRO","trust","zero trust","architecture","verification","segmentation","access","configure","critical","implementing",{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":43},"01M6Q0F1YZJW3KX5FBKEVZ8CV4","zcode-delegate","Delegate coding tasks to the Z.AI ZCode CLI only when the user explicitly requests it, while the orchestrator retains review and landing responsibility.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fzcode-delegate\u002FSKILL.md",[],[103,104,105,106,107,108,109,110,111,36,112,113,114,35,115,34,116,117,94,118],"delegate","zcode","zcode delegate","review","cli","orchestration","tasks","run","Bash","node","read","orchestrator","files","desktop app","app","agent-orchestration",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":43},"01M6Q0ETK41VGNMYEXW0T00AJN","youtube-transcript","Fetch YouTube transcripts through DeepAPI or local fallback tooling and save clean text output.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-transcript\u002FSKILL.md",[],[126,127,128,41,129,130,131,132,133,134,135,136,137,34,33,138,139,140,141,142],"transcript","youtube transcript","youtube","github","codex","research","license","https","text","fetch","tooling","clean","local","output","deepapi","youtube transcripts","blob",{"skillNo":144,"name":145,"nameZh":6,"description":146,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":147,"categories":148,"keywords":160,"publishedAt":43},"01M6Q0EPPH2PVMDW40017F8WX0","youtube-transcript-skills","Fetch YouTube video transcripts, search videos\u002Fchannels, browse channels, and extract playlists via the getyoutubetranscript.com API - free tier, no card required.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-transcript-skills\u002FSKILL.md",[149,152,155],{"categoryNo":150,"categoryName":151,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000010","后端与接口",{"categoryNo":153,"categoryName":154,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000016","图像与多媒体",{"categoryNo":156,"categoryName":157,"taxonomyType":158,"taxonomyTypeDesc":159},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[126,127,128,161,41,162,129,130,132,163,164,165,166,135,167,34,33,168,169,170],"youtube transcript skills","gemini","search","api","Video","browse","api-integration","free","card","channels",{"skillNo":172,"name":173,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":181,"publishedAt":43},"01M6Q0EAB96PX83W5PTE8A0R8D","youtube-notetaker","Turn YouTube talks into local study notes with slides, transcripts, editable annotations, and a markdown-backed viewer.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-notetaker\u002FSKILL.md",[177,180],{"categoryNo":178,"categoryName":179,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":153,"categoryName":154,"taxonomyType":20,"taxonomyTypeDesc":21},[128,182,183,107,41,184,185,186,187,132,165,188,189,190,191,33,192,138,193,194],"youtube notetaker","notetaker","workflow","official","notes","code","markdown","slides","annotations","README","study","dair-academy","viewer",{"skillNo":196,"name":197,"nameZh":6,"description":198,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":199,"categories":200,"keywords":202,"publishedAt":43},"01M6Q0E55F0GW5MX48V8HXS8B5","youtube-full","Fetch YouTube transcripts, search videos, browse channels, and extract playlists via TranscriptAPI — no yt-dlp, no Google API key, works from any cloud server.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-full\u002FSKILL.md",[201],{"categoryNo":150,"categoryName":151,"taxonomyType":20,"taxonomyTypeDesc":21},[203,128,204,205,41,162,163,206,207,164,165,166,135,208,167,34,33,168,209,141],"youtube full","full","google","server","dlp","credits","automatic"]