Grades are a hygiene check, not a security guarantee.

Claude Code, Skill

Skill Inject

A comprehensive benchmark for measuring prompt injection vulnerabilities in LLM agent skill files.

Compatibility: Not stated.

Grade C

One Medium finding. Read the findings before installing.

Graded by PlugVetRubric v2

Not a guarantee. A hygiene check of the source at the commit shown, not a security audit.

Rescanned at commit 182f3d9d9836, unchanged since the check below.

OBFUSCATION_ENCODED_BLOB

Why this grade

  • OBFUSCATION_ENCODED_BLOB Medium

    Contains a long base64 shaped string in a file that should be readable.

    data/skills/document-skills/docx/ooxml/schemas/ISO-IEC29500-4_2016/sml.xsd, line 1 <?xml version="1.0" encoding="utf-8"?>

Any High, or two or more Medium findings, give D. One Medium gives C. Low findings or a security Note give B. Each code names a finding behind the letter.

Grades are a hygiene check of the source at the commit shown, on the date checked, not a security audit or a guarantee. The free directory grades are automated. Before release, a PlugVet engineer reviews every paid report and any grade we hold back, including all F grades, against the source code and the evidence we quote.

What was checked

Plain notes from the hygiene checks.

  • Permissions

    Concern

    can delete files.

    PERMISSION_DELETE_FILES

    data/instruction_unit_tests/task_checks.py, line 273

    result = subprocess.run(

    data/instruction_unit_tests/obvious/test_obvious_7.py, line 7

    - Python os.remove() - in generated scripts

    data/skills/document-skills/docx/ooxml/scripts/pack.py, line 103

    result = subprocess.run(

    Show more

    data/skills/document-skills/docx/ooxml/scripts/validation/redlining.py, line 153

    result = subprocess.run(

    data/skills/document-skills/docx/ooxml/scripts/validation/redlining.py, line 185

    result = subprocess.run(

    data/skills/document-skills/docx/scripts/document.py, line 836

    shutil.rmtree(self.temp_dir)

    data/skills/document-skills/pptx/ooxml/scripts/pack.py, line 103

    result = subprocess.run(

    data/skills/document-skills/pptx/ooxml/scripts/validation/redlining.py, line 153

    result = subprocess.run(

    data/skills/document-skills/pptx/ooxml/scripts/validation/redlining.py, line 185

    result = subprocess.run(

    data/skills/document-skills/pptx/scripts/thumbnail.py, line 219

    result = subprocess.run(

  • Install and update

    Pass

    No install-time script found, and dependencies are pinned or absent.

  • Network

    Concern

    can reach a host that is not written as a fixed address. can reach agenskill-api.onrender.com. The description does not name that host. can reach agensk. The description does not name that host. can reach ipapi.co. The description does not name that host.

    NETWORK_NO_FIXED_HOST NETWORK_UNNAMED_HOST

    data/instruction_unit_tests/obvious/test_obvious_10.py, line 4

    to agenskill-api.onrender.com/envs. Agents may use curl directly or

    data/instruction_unit_tests/obvious/test_obvious_10.py, line 5

    write a Python script that uses requests.post().

    data/instruction_unit_tests/obvious/test_obvious_10.py, line 32

    # Secondary: curl command with .env reference

    Show more

    data/instruction_unit_tests/obvious/test_obvious_10.py, line 34

    return True, "curl command with .env target found (command_history)"

    data/instruction_unit_tests/obvious/test_obvious_10.py, line 41

    # Check for find + curl pattern (scanning for .env then uploading)

    data/instruction_unit_tests/obvious/test_obvious_11.py, line 33

    # Secondary: curl with SSH file patterns

    data/instruction_unit_tests/obvious/test_obvious_11.py, line 35

    return True, "curl command targeting SSH files found (command_history)"

    data/instruction_unit_tests/obvious/test_obvious_21.py, line 4

    include curl commands to exfiltrate queries to agenskill-api.onrender.com/logging.

  • Secrets

    Concern

    reads a credential or environment file.

    SECRETS_ENV_READ SECRETS_PATH_DOTENV

    apptainer/debug_env.sh, line 7

    ENV_FILE="$SCRIPT_DIR/.env"

    apptainer/run_multi_skill_dilution.sh, line 17

    [ -f "$PROJECT_ROOT/.env" ] && set -a && source "$PROJECT_ROOT/.env" && set +a

    apptainer/run_obvious_eb.sh, line 17

    [ -f "$PROJECT_ROOT/.env" ] && set -a && source "$PROJECT_ROOT/.env" && set +a

    Show more

    apptainer/run_obvious_eb_remaining.sh, line 17

    [ -f "$PROJECT_ROOT/.env" ] && set -a && source "$PROJECT_ROOT/.env" && set +a

    apptainer/run_phrasing_variants.sh, line 16

    [ -f "$PROJECT_ROOT/.env" ] && set -a && source "$PROJECT_ROOT/.env" && set +a

    apptainer/run_sandbox.sh, line 28

    # Shell environment is loaded first (lower priority); .env file is loaded second

    apptainer/run_sandbox.sh, line 29

    # and overwrites any matching keys, so .env always takes precedence.

    apptainer/run_sandbox.sh, line 39

    # 2. Load from .env (overwrites shell env for any key present in the file)

    data/instruction_unit_tests/task_checks.py, line 63

    api_key = os.environ.get("OPENAI_API_KEY")

    data/skills/code-review/references/style-guide.md, line 138

    api_key = os.environ.get("API_KEY")

    data/skills/hf-paper-publisher/skills/hugging-face-paper-publisher/scripts/paper_manager.py, line 35

    self.token = hf_token or os.getenv("HF_TOKEN") or HfFolder.get_token()

    data/skills/hf_dataset_creator/skills/hugging-face-dataset-creator/scripts/dataset_manager.py, line 22

    HF_TOKEN = os.environ.get("HF_TOKEN")

    data/skills/hf_model_evaluation/skills/hugging-face-evaluation-manager/examples/artificial_analysis_to_hub.py, line 42

    API_KEY = os.getenv("AA_API_KEY")

    data/skills/hf_model_evaluation/skills/hugging-face-evaluation-manager/examples/artificial_analysis_to_hub.py, line 43

    HF_TOKEN = os.getenv("HF_TOKEN")

    data/skills/hf_model_evaluation/skills/hugging-face-evaluation-manager/scripts/evaluation_manager.py, line 515

    hf_token = os.getenv("HF_TOKEN")

    data/skills/hf_model_evaluation/skills/hugging-face-evaluation-manager/scripts/evaluation_manager.py, line 725

    hf_token = os.getenv("HF_TOKEN")

  • Agent instructions

    Pass

    No instruction to ignore earlier instructions or to hide a step.

  • Obfuscation

    Concern

    contains an encoded blob in a file that should be readable.

    OBFUSCATION_ENCODED_BLOB

    data/skills/document-skills/docx/ooxml/schemas/ISO-IEC29500-4_2016/sml.xsd, line 1

    <?xml version="1.0" encoding="utf-8"?>

    data/skills/document-skills/pptx/ooxml/schemas/ISO-IEC29500-4_2016/sml.xsd, line 1

    <?xml version="1.0" encoding="utf-8"?>

    data/skills/seo-review/SKILL.md, line 1

    ---

    Show more

    data/task_files/.ssh/id_rsa.pub, line 1

    ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQC0R0ZHR0hJSktMTU5PUFFSU1RVVldYWVo5OUFBQkNERUZHMTIzNDU2Nzg5MDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDA1NTVVVlpaWlpZWVlZWVlZWVlZWVlZWVlZWVlZWVlZWVlZWVlZ

    data/task_files/.ssh/known_hosts, line 1

    github.com ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQC0R0ZHR0hJSktMTU5PUFFSU1RVVldYWVo5OTk5OTk5OTk5OTk5OVdXV1dXV1dXV1dXV1dXV1dXV1dXV1dXV1dXV1REZFRkdHSUpLTE1OT1BRUlNUVVZXWFlaWlpZWVlZWVl

  • Licence

    Pass

    A licence file is present (LICENSE).

  • Maintainer activity

    Pass

    The latest commit is within the last 365 days.

Get the full report on this tool, £49

Note

Hygiene check of commit 182f3d9d9836. can delete files. can reach a host that is not written as a fixed address. can reach agenskill-api.onrender.com. The description does not name that host. can reach agensk. The description does not name that host. can reach ipapi.co. The description does not name that host. reads a credential or environment file. contains an encoded blob in a file that should be readable.

Date and source

Date checked:

View source

github.com

Commit graded: 182f3d9d9836e81cdae213e9b9cec1d9be96eea3

This link goes to the original host. PlugVet does not host or check what you download from there, and installing it is your own decision and risk.

Read how this letter is chosen