Back to skill

Security audit

YouCam Beauty Advisor

Security checks for vulnerabilities and agentic risk

Overview

The skill’s makeup workflow matches its stated purpose, with expected selfie processing by YouCam and user-approved credit costs.

Use this skill only with photos you are comfortable sending to YouCam/Perfect Corp, and review the provider's privacy and retention terms before uploading. Confirm credit costs and the selected preview option. Protect your dedicated API key and use an isolated environment with reviewed dependency versions. This assessment reviewed the packaged source; it did not verify live service behavior or installed dependency versions.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
Findings (28)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The code exposes a general YouCam API interface rather than a makeup-scoped implementation. The run command forwards the supplied feature and parameters to the task runner after loading that feature's contract, without enforcing the description's exclusions. Skin analysis is shown in executable CLI examples, and hair/clothing workflows are explicitly accommodated by reference-image arguments. The supplied chunk does not itself implement face-shape or skin-tone-based makeup recommendations. Setup, authentication, and credit checks are supporting details, not independent reasons for the mismatch. The exact set of executable features depends on imported modules and contracts that were not supplied.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The feature argument is passed to contract loading, task creation, and polling without a makeup-only allowlist. Reference inputs are also supported generically, including for hair and garment workflows. Thus, the supplied code exposes broader processing capabilities than the declared makeup-only scope. Exact available features depend on the unseen API contracts. Uploading images, downloading remote inputs, using temporary files, and polling tasks are supporting implementation details and are not independently treated as mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

The API helpers accept a caller-selected feature and resolve its endpoint from an external contract rather than restricting operations to makeup try-on. The code explicitly references skin-analysis and hair-transfer, which fall outside the declared scope. This supports a broader API capability than the description represents, although the supplied chunk does not demonstrate an actual skin or hair processing invocation, and the available features depend on the unseen contract. Credential loading, uploads, pricing, and credit checks are ordinary supporting implementation details, not independent mismatches. The chunk does not itself implement face analysis or makeup recommendation, but those steps could exist elsewhere.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 12)May include surrounding context.

md
requirements:
  credentials:
    - name: YOUCAM_API_KEY
      source: env | credentials.json (in the skill root, next to SKILL.md)
  network:
    - yce-api-01.makeupar.com
  apis:

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 54)May include surrounding context.

md
1. A clear, front-facing selfie.
2. Run `python scripts/youcam_core.py validate-key` (0 credits) to confirm the key. If it is not set,
   tell the user to get one at <https://yce.perfectcorp.com/api-console> and set the
   `YOUCAM_API_KEY` env var, or create `credentials.json` in the skill root (see
   `credentials.example.json` for the format).

## Flow

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 · SKILL.md (reported line 82)May include surrounding context.

md
using the `task_id` from that output, until `task_status` is `success` or `error`. A task_id
   stays valid for 24 hours.

## Output rules
* **Return the result image URL exactly as given** (you may make it a hyperlink, but do not
  modify or shorten it).
* Explain the recommendation (why this look suits their face shape / skin tone).

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/constants.py (reported line 7)May include surrounding context.

python
skills/<skill-name>/
  ├── SKILL.md
  ├── credentials.example.json   # template for credentials.json
  ├── credentials.json           # CREDENTIALS_FILE (user-provided)
  └── scripts/                   # SKILL_SCRIPT_ROOT (this file's folder)
      ├── constants.py, youcam_api.py, ...

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/constants.py (reported line 8)May include surrounding context.

python
skills/<skill-name>/
  ├── SKILL.md
  ├── credentials.example.json   # template for credentials.json
  ├── credentials.json           # CREDENTIALS_FILE (user-provided)
  └── scripts/                   # SKILL_SCRIPT_ROOT (this file's folder)
      ├── constants.py, youcam_api.py, ...
      ├── api-fallback.yaml      # API_FALLBACK_DOC

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/constants.py (reported line 15)May include surrounding context.

python
├── requirements.txt       # REQUIREMENTS_FILE
      └── .venv/                 # VENV_DIR

Every path the scripts use resolves inside `SKILL_SCRIPT_ROOT`, except credentials.json, which lives
one level up in the skill root so users edit it next to SKILL.md rather than inside the vendored code.
"""

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/constants.py (reported line 23)May include surrounding context.

python
SKILL_SCRIPT_ROOT = os.path.dirname(os.path.abspath(__file__))
API_FALLBACK_DOC = os.path.join(SKILL_SCRIPT_ROOT, "api-fallback.yaml")
CREDENTIALS_FILE = os.path.join(os.path.dirname(SKILL_SCRIPT_ROOT), "credentials.json")

# Install dependencies into a local virtual environment in the skill script folder.
VENV_DIR = os.path.join(SKILL_SCRIPT_ROOT, ".venv")

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/youcam_api.py (reported line 90)May include surrounding context.

python
def get_key() -> str:
    """Resolve the API key from the YOUCAM_API_KEY env var or credentials.json.

    Exits with code 2 and a structured error if no usable key is configured.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/youcam_api.py (reported line 108)May include surrounding context.

python
"No API key configured.",
            hint=f'Set the YOUCAM_API_KEY env var, or create {CREDENTIALS_FILE} with {{"api_key": "..."}}'
            + format_hint,
            details="Neither YOUCAM_API_KEY nor credentials.json was found.",
            code=2,
        )

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/youcam_core.py (reported line 9)May include surrounding context.

python
everything the skill needs.

Auth:
    Set the YOUCAM_API_KEY environment variable, or create a credentials.json file (in the
    skill's root, next to SKILL.md) containing {"api_key": "..."}. The env var is used first
    if both are present. If neither is set, the command & API call fails.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/youcam_tasks.py (reported line 123)May include surrounding context.

python
try-on. Follows the same rules as src_file. Not every feature needs this.
        params: Extra options to send along with the request.
        key: The API key to use. Required; the caller (e.g. youcam_core.py's CLI) is
            expected to resolve it from YOUCAM_API_KEY or credentials.json beforehand.
        version: Which version of the feature to call. Defaults to the standard version.
        interval: How many seconds to wait between checks while the job is processing.
        max_polls: How many times to check on the job before giving up and returning

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

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

Content

Scanner excerpt · scripts/youcam_tasks.py (reported line 141)May include surrounding context.

python
contract = contract or load_contract(feature)
    if not key:
        raise ValueError(
            "missing API key. set the YOUCAM_API_KEY env var, or create credentials.json "
            "(see credentials.example.json for the format)"
        )
    body = dict(params or {})

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The workflow sends the user's selfie to an external service without an explicit privacy disclosure.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow instructs the agent to submit a user's identifiable selfie to an external analysis and rendering service without requiring a user-facing disclosure of that transfer. Naming YouCam in the skill metadata and obtaining approval for credit costs do not explicitly establish informed consent to third-party processing of facial imagery.

Content

No source excerpt is available for this finding.

Tainted flow: 'put_req' from requests.post (line 295, network input) → requests.put (network output)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/youcam_api.py (reported line 297)May include surrounding context.

python
info = meta.json()["data"]["files"][0]
    put_req = info["requests"][0]
    with open(file_path, "rb") as fh:
        requests.put(
            put_req["url"],
            headers=put_req.get("headers", {"Content-Type": ct}),
            data=fh,

The task API exposes unrestricted YouCam feature execution despite the skill's makeup-only scope.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest explicitly excludes skin-condition scoring, hair, and clothes, but create_task accepts any contract-defined feature and submits its payload without a makeup-only allowlist (L306-L324); its documentation even names skin-analysis as an example. The same generic interface supports hair-transfer template retrieval (L209-L247), exposing capabilities beyond those justified by makeup recommendation and try-on, although this file does not establish that those features are actually invoked.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/youcam_api.py (reported line 313)May include surrounding context.

python
Args:
        contract: The feature's loaded API contract, e.g. from load_contract().
        feature: Which YouCam feature to create a task for, e.g. "skin-analysis".
        body: The request payload, e.g. file_id(s)/url(s) and any feature-specific params.
        key: The API key to authenticate the request with.
        version: Which version of the feature to call. Defaults to the standard version.

The makeup-only skill exposes a general-purpose YouCam feature runner without enforcing the manifest's exclusions.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest explicitly excludes skin-condition scoring, hair, and clothes, but L203-L216 load and execute the caller-supplied feature without a makeup-specific scope check. This broader scope is corroborated by the skin-analysis execution example at L019-L020 and the hair-transfer/clothing reference-image option at L141-L143.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/youcam_init.py (reported line 49)May include surrounding context.

python
def _run_setup_step(cmd: list[str]) -> None:
    """Run one setup command, exiting with a structured error if it fails."""
    try:
        subprocess.run(cmd, check=True)
    except subprocess.CalledProcessError as exc:
        exit_with_error(
            "setup_failed",

The task runner supports arbitrary YouCam features, including categories explicitly excluded by the beauty-advisor manifest.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest limits the skill to makeup recommendation and preview and explicitly excludes skin-condition scoring, hair, and clothes. The runner instead loads a contract for the caller-provided feature and submits it without a makeup-only restriction; its documentation explicitly identifies skin-analysis (L112), hair transfer, and garment try-on (L118–L120) as supported uses.

Content

No source excerpt is available for this finding.

The mandatory closing note forces English text without user language choice.

Low
Category
Output Handling
Confidence
92% confidence
Finding

The instruction requires every response to end with a fixed English disclaimer, regardless of the user's language. It provides neither a translation option nor a justification for requiring English.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
92% confidence
Finding

The open-ended requests requirement permits installations to resolve different, unreviewed releases and does not authenticate package artifacts with hashes. This is a supply-chain and reproducibility weakness, not evidence of a compromised dependency or malicious intent.

Content

Scanner excerpt · scripts/requirements.txt (reported line 1)May include surrounding context.

text
requests>=2.28
PyYAML>=6.0