Back to skill

Security audit

flightroutes24 ai

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent flight-search and booking integration, but it stores sensitive booking data locally and ships unrelated agent permissions that deserve manual review before installation.

Review this skill before installing. Use it only in a workspace where local .cache data is protected, do not share or commit the skill directory after entering credentials or passenger details, remove the bundled .claude Git permissions, and avoid production booking unless you understand the FR24 account and gateway controls.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (6)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/config_keys.py:71
Finding

Procurement credentials are stored in plaintext without restrictive file permissions

Content
View full analysis
dict: CACHE_DIR.mkdir(parents=True, exist_ok=True) existing: dict = {} if KEYS_FILE.is_file(): try: existing = json.loads(KEYS_FILE.read_text(encoding="utf-8")) except (json.JSONDecodeError, OSError): pass field_map = { "appkey": "appkey", "sign-secret": "signSecret", "sign_secret": "signSecret", "aes-secret": "aesSecret", "aes_secret": "aesSecret", } updated = [] for key, val in values.items(): json_key = field_map.get(key, key) if val: existing[json_key] = val updated.append(json_key) KEYS_FILE.write_text( json.dumps(existing, ensure_ascii=False, indent=2), encoding="utf-8", ) ``` The CLI also accepts secrets directly as command-line arguments: ```python p_set.add_argument("--appkey", default="", help="Procurement APPKEY") p_set.add_argument("--sign-secret", default="", help="SHA512 signing secret") p_set.add_argument("--aes-secret", default="", help="16-byte AES secret") ``` ### Technical Analysis The Skill writes the procurement APPKEY, signing secret, and passenger-encryption key to `.cache/keys.json` as unencrypted JSON. It does not explicitly create the directory with a restrictive mode, create the credential file with mode `0600`, verify ownership, or protect against symbolic-link replacement. Passing the secrets through command-line arguments can additionally expose them through shell history, process inspection, command telemetry, or Agent execution logs. The reviewed project tree did not contain the `.gitignore` file that the documentation claims protects `.cache` and `.env` files. Possession of all three values can enable pr ...[truncated 1137 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/skill_booking_client.py:53
Finding

Passenger identity and contact records are retained in plaintext cache files

Content
View full analysis
None: path.parent.mkdir(parents=True, exist_ok=True) path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8") ``` ```python _save_json( PASSENGERS_FILE, { "passengers": passengers, "agentContact": contact, "passengerRawMappings": raw_mappings, "contactRaw": contact_raw, }, ) ``` ### Technical Analysis The passenger parsing flow writes structured passenger data, contact information, raw field mappings, and raw contact input to `.cache/passengers.json`. Based on the surrounding parser and test flow, this data can include names, birth dates, passport numbers and expiry dates, nationality, phone numbers, and email addresses. The file is not encrypted at rest, no restrictive file mode is enforced, and the code does not delete it after verification, successful booking, failure, or cancellation. Retaining both normalized fields and raw mappings duplicates sensitive data and increases the disclosure surface. The network-facing passenger field is encrypted before booking, but that protection does not apply to this local cache file. ### Attack Path 1. A user supplies passenger and contact details for a booking. 2. `cmd_parse_passengers` parses the input and writes the complete record to `.cache/passengers.json`. 3. The file remains after the booking flow finishes or is abandoned. 4. Another local process, local user, backup tool, synchronization service, malicious Agent command, or accidental repository commit reads the file. 5. The attacker extracts reusable identity and contact information. ### Impact Assessment Successful exploitation exposes high-value personally identifiable information. The affected scope ca ...[truncated 331 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/newapi_auth.py:27
Finding

Passenger encryption uses a fixed IV and provides no ciphertext authentication

Content
View full analysis
str: """ AES/CBC/PKCS5Padding, IV is 16 zero bytes; key must be exactly 16 bytes when UTF-8 encoded. """ key = password.encode("utf-8") if len(key) != 16: raise ValueError( f"NewApi AES key must be 16 bytes; current length is {len(key)}." "Check FR_NEWAPI_AES_SECRET." ) cipher = AES.new(key, AES.MODE_CBC, iv=b"\x00" * 16) encrypted = cipher.encrypt(pad(content.encode("utf-8"), AES.block_size)) return base64.b64encode(encrypted).decode("ascii") ``` ### Technical Analysis AES-CBC requires an unpredictable and unique IV for each encryption operation. The implementation instead uses an all-zero IV for every passenger payload. As a result, identical plaintext prefixes encrypted under the same key produce identical ciphertext blocks, allowing cross-request correlation. CBC also does not provide integrity or authenticity. Without a message authentication code, an attacker capable of modifying ciphertext can perform controlled bit-flipping operations. Whether those changes are exploitable depends on server-side parsing and validation. Base64 encoding is only a transport representation and does not add confidentiality or integrity. The pre-scan encoding alert therefore reflects declared passenger encryption rather than an independent covert-exfiltration mechanism. ### Attack Path 1. An attacker obtains multiple encrypted booking payloads through compromised TLS termination, gateway logging, proxy access, or server-side telemetry. 2. Because the same key and zero IV are reused, the attacker compares ciphertext blocks and correlates repeated passenger-data prefixes. 3. If the attacker can modify a request in transit or before proces ...[truncated 687 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/newapi_client.py:108
Finding

Distributed client can assert authentication and IP-whitelist bypass headers

Content
View full analysis
dict[str, str]: headers = { "Content-Type": "application/json; charset=utf-8", "appkey": NEWAPI_APP_KEY, "Accept-Encoding": "gzip", } if GRAY_HEADER: headers["gray"] = GRAY_HEADER if NEWAPI_SKIP_AUTH: headers["fr24-skip-auth"] = "1" return headers def _newapi_headers() -> dict[str, str]: headers = _newapi_headers_base() if NEWAPI_SKIP_IP_WHITELIST: headers[FR24_API_HEADER] = "1" return headers def _newapi_channel_headers() -> dict[str, str]: """NewApi headers for pricing and order write operations.""" headers = _newapi_headers_base() headers[FR24_API_HEADER] = "1" return headers ``` The flags are loaded from local configuration in `config.py:44-45, 124-126`: ```python "FR_NEWAPI_SKIP_AUTH": "skipAuth", "FR_NEWAPI_SKIP_IP_WHITELIST": "skipIpWhitelist", ... NEWAPI_SKIP_AUTH = _read_config_bool("FR_NEWAPI_SKIP_AUTH") NEWAPI_SKIP_IP_WHITELIST = _read_config_bool("FR_NEWAPI_SKIP_IP_WHITELIST") FR24_API_HEADER = "fr24-api" ``` ### Technical Analysis The distributed Skill allows local environment or cache configuration to add a header explicitly named `fr24-skip-auth`. It also adds `fr24-api: 1` to bypass the IP whitelist when configured, and booking-channel requests add `fr24-api: 1` unconditionally. ...[truncated 1565 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
.claude/settings.local.json:1
Finding

Skill package pre-authorizes unrelated Git remote and push operations

Content
View full analysis
Remediation
View remediation

other

Warning
Location
scripts/nl_scenario_test.py:142
Finding

Production-facing scenario test rotates client identities after quota exhaustion

Content
View full analysis
int: CACHE.mkdir(parents=True, exist_ok=True) report: list[dict] = [] search_key = secrets.token_urlsafe(32) for sc in SCENARIOS: row: dict = {"id": sc.id, "note": sc.note, "text": sc.text or "(intent)"} pr = run_parse(sc) row["parse"] = pr parse_ok = pr["ok"] expect_ok = sc.expect_parse == "success" row["parseMatch"] = parse_ok == expect_ok if parse_ok and sc.expect_search != "skip": sr = run_search(pr["payload"], search_key) row["search"] = sr if sc.expect_search == "success": row["searchMatch"] = sr["ok"] elif sc.expect_search == "fail": row["searchMatch"] = not sr["ok"] else: row["searchMatch"] = True if sr.get("code") in ("307901", "404"): row["searchNote"] = f"infra/code={sr.get('code')}, rotated client key" search_key = secrets.token_urlsafe(32) ``` ### Technical Analysis The test sends requests to the configured shopping gateway using a generated client identity. When the service reports code `307901`, documented as demo-quota exhaustion, the test replaces the client key and continues. A per-client daily quota is intended to limit resource consumption. Generating a new identity after exhaustion defeats that accounting model. Including HTTP 404 in the same rotation behavior is also inappropriate because changing identity does not correct a missing endpoint. The transmitted scenario payloads are predefined itinerary data rather than credentials or passenger identity records. Therefore, the pre-scan network alert is not evidence of covert sensitive-data exfiltration; the confirmed issue is quota evasion. ### Attack Path ...[truncated 881 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (81)

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · README.en.md (reported line 140)May include surrounding context.

md
| [INSTALL.md](./INSTALL.md) | Installation |
| [references/booking.md](./references/booking.md) | Booking steps |
| [references/search_params.md](./references/search_params.md) | Search request fields |
| [references/output-rules.md](./references/output-rules.md) | User-visible output rules |
| [references/user-appkey-config.md](./references/user-appkey-config.md) | Procurement key setup (users) |
| [references/setup-maintainer.md](./references/setup-maintainer.md) | Maintainer notes (not for end users) |

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented behavior extends beyond simple fare lookup into local secret management, environment inspection, and persistent key storage/clearing. When a skill's declared purpose understates credential-handling and local state mutation, users and orchestrators may invoke it without understanding the security implications, which can lead to mishandling of procurement secrets or unsafe automation of sensitive operations.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 15)May include surrounding context.

python
BOOKING_CONTEXT_FILE = CACHE_DIR / "booking_context.json"
PASSENGERS_FILE = CACHE_DIR / "passengers.json"
KEYS_FILE = CACHE_DIR / "keys.json"
ENV_FILE = SKILL_DIR / ".env"

CLIENT_KEY_HEADER = "X-Skill-Client-Key"
DAILY_LIMIT = 10

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 37)May include surrounding context.

python
# ---------------------------------------------------------------------------
# 多源配置读取:环境变量 > .env 文件 > .cache/keys.json
# ---------------------------------------------------------------------------

_ENV_TO_JSON_KEYS = {

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 53)May include surrounding context.

python
# ---------------------------------------------------------------------------
# 多源配置读取:环境变量 > .env 文件 > .cache/keys.json
# ---------------------------------------------------------------------------

_ENV_TO_JSON_KEYS = {

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 91)May include surrounding context.

python
# ---------------------------------------------------------------------------
# 多源配置读取:环境变量 > .env 文件 > .cache/keys.json
# ---------------------------------------------------------------------------

_ENV_TO_JSON_KEYS = {

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 96)May include surrounding context.

python
# ---------------------------------------------------------------------------
# 多源配置读取:环境变量 > .env 文件 > .cache/keys.json
# ---------------------------------------------------------------------------

_ENV_TO_JSON_KEYS = {

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · config.py (reported line 114)May include surrounding context.

python
# ---------------------------------------------------------------------------
# 多源配置读取:环境变量 > .env 文件 > .cache/keys.json
# ---------------------------------------------------------------------------

_ENV_TO_JSON_KEYS = {

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/booking_flow_test.py (reported line 14)May include surrounding context.

python
_ROOT = Path(__file__).resolve().parent.parent
_SCRIPTS = Path(__file__).resolve().parent
PY = sys.executable
ENV = {**os.environ, "PYTHONIOENCODING": "utf-8"}

QUERY = os.environ.get("FR_BOOKING_TEST_QUERY", "深圳到曼谷 6月2日")
PAX_TEXT = os.environ.get(

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/validate_user_output.py (reported line 92)May include surrounding context.

python
_ROOT = Path(__file__).resolve().parent.parent
_SCRIPTS = Path(__file__).resolve().parent
PY = sys.executable
ENV = {**os.environ, "PYTHONIOENCODING": "utf-8"}

QUERY = os.environ.get("FR_BOOKING_TEST_QUERY", "深圳到曼谷 6月2日")
PAX_TEXT = os.environ.get(

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The installation guide says the skill may be auto-matched based on the SKILL.md description with examples like '查航班、搜机票、预订等'. These phrases are common user intents and the document does not provide negative examples or scope boundaries, which could lead to unintended invocation in general travel-related conversation.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises and documents access to sensitive capabilities including environment-based secret retrieval, local file reads/writes, shell command execution, and network requests, but does not declare any tool or permission scope. That creates an over-privileged integration surface where an agent runtime may allow more capability than users or reviewers expect, increasing the risk of secret exposure, unintended command execution, or unsafe external actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description states that the skill supports booking and order creation for real flight purchases, but the upfront description does not prominently warn that actions may create real orders with external impact. In this context, weak user-facing signaling is dangerous because mistaken invocation or misunderstanding could lead to irreversible commercial actions, charges, or personal data submission.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger list includes the very broad everyday term “飞” (“fly”), which can match many benign conversations unrelated to ticket purchase. In a skill that can search, validate, and create real bookings, accidental invocation is risky because it can initiate quota-consuming searches or progress toward system-impacting actions without sufficiently intentional user activation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module description and subsequent user-facing strings are written only in Chinese, indicating the skill communicates in a fixed language by default. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless a justified locale restriction is documented, which is not present here.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
91% confidence
Finding

The JSON key mapping for skipAuth mirrors the same insecure bypass in an alternate config source, increasing the chance of accidental enablement from local cached configuration. Supporting the bypass in multiple secret sources broadens the attack surface for misconfiguration and unauthorized booking/pricing access.

Content

Scanner excerpt · config.py (reported line 44)May include surrounding context.

python
"FR_NEWAPI_APPKEY": "appkey",
    "FR_NEWAPI_SIGN_SECRET": "signSecret",
    "FR_NEWAPI_AES_SECRET": "aesSecret",
    "FR_NEWAPI_SKIP_AUTH": "skipAuth",
    "FR_NEWAPI_SKIP_IP_WHITELIST": "skipIpWhitelist",
}

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
91% confidence
Finding

The JSON key mapping for skipAuth mirrors the same insecure bypass in an alternate config source, increasing the chance of accidental enablement from local cached configuration. Supporting the bypass in multiple secret sources broadens the attack surface for misconfiguration and unauthorized booking/pricing access.

Content

Scanner excerpt · config.py (reported line 44)May include surrounding context.

python
"FR_NEWAPI_APPKEY": "appkey",
    "FR_NEWAPI_SIGN_SECRET": "signSecret",
    "FR_NEWAPI_AES_SECRET": "aesSecret",
    "FR_NEWAPI_SKIP_AUTH": "skipAuth",
    "FR_NEWAPI_SKIP_IP_WHITELIST": "skipIpWhitelist",
}

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
89% confidence
Finding

At module load time, the code reads FR_NEWAPI_SKIP_AUTH and FR_NEWAPI_SKIP_IP_WHITELIST into active configuration, making insecure toggles first-class runtime behavior. In the context of a booking/procurement integration, these flags can disable core protections and enable unauthorized API use if an attacker or operator can alter environment or local config.

Content

Scanner excerpt · config.py (reported line 124)May include surrounding context.

python
NEWAPI_APP_KEY = _read_config("FR_NEWAPI_APPKEY")
NEWAPI_SIGN_SECRET = _read_config("FR_NEWAPI_SIGN_SECRET")
NEWAPI_AES_SECRET = _read_config("FR_NEWAPI_AES_SECRET")
NEWAPI_SKIP_AUTH = _read_config_bool("FR_NEWAPI_SKIP_AUTH")
NEWAPI_SKIP_IP_WHITELIST = _read_config_bool("FR_NEWAPI_SKIP_IP_WHITELIST")
FR24_API_HEADER = "fr24-api"

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The strings in USER_BOOKING_USER_MESSAGE, SEARCH_ONLY_HINT, and USER_SKILL_QUOTA_EXCEEDED_MESSAGE are end-user communications and are all hardcoded in Chinese. There is no indication of user opt-in, alternative locale support, or documented justification for restricting the skill to this language.

Content

No source excerpt is available for this finding.