Bawwab: Top 302 scan

huggingface-skills

Not malicious, needs a look Reviewed: expected behaviour

Bawwab's static rules found patterns that deserve a human look. This is not a finding that the package is malicious; the review notes below say what each pattern turned out to be.

Why it was flagged

The static rules marked this “Not malicious, needs a look” because of:

Kind
Plugin
Source
github
Repository
huggingface/skills
Commit
c3ff942be3e07116b475253f0a8fc4f54682c1a8
Listed in
anthropics/claude-plugins-official
Popularity
11,366 GitHub stars (2026-11-20) (official, always included)
Link
https://github.com/huggingface/skills/tree/c3ff942be3e07116b475253f0a8fc4f54682c1a8
Scanned
2026-11-20T23:14:16+00:00
Scan time
6.4 s
Scanned as
plugin
Review note: A CI step installs the vendor's own CLI, and a skill tells the agent to cd before running scripts; nothing malicious.

Findings (2)

remote-code.pipe-shell: Pipes a download into a shell high

.github/workflows/publish-single-clawhub-skill.yml:113

OWASP LLM LLM03 Supply ChainOWASP Agentic ASI04 Agentic Supply Chain VulnerabilitiesOWASP Agentic ASI05 Unexpected Code Execution (RCE)MITRE ATLAS AML.T0050 Command and Scripting InterpreterCWE CWE-494 Download of Code Without Integrity Check

run: curl -LsSf https://hf.co/cli/install.sh | bash -s

What this means: Executes whatever the remote server returns, unreviewed.

The rule's general advice: Download, verify a checksum, then run. (see the review note for this case)

Expected for what this tool does

A GitHub Actions step on the maintainers' runner that installs the Hugging Face CLI from hf.co/cli/install.sh. It is the vendor's own installer and never runs on users' machines.

Reviewed by Claude (AI-assisted review), 2026-11-21

injection.covert-action: Tells the AI to read or send sensitive data high

skills/huggingface-paper-publisher/SKILL.md:66

OWASP LLM LLM01 Prompt InjectionOWASP LLM LLM02 Sensitive Information DisclosureOWASP Agentic ASI01 Agent Goal HijackOWASP Agentic ASI02 Tool Misuse and ExploitationMITRE ATLAS AML.T0051.001 LLM Prompt Injection: IndirectMITRE ATLAS AML.T0110.000 AI Agent Tool Poisoning: Definition and InstructionsMITRE ATLAS AML.T0086 Exfiltration via AI Agent Tool InvocationCWE CWE-1427 Improper Neutralization of Input Used for LLM Prompting

Before running any script, first

What this means: Instructions to read keys or secrets, include them in replies or send them somewhere, or to do a hidden step first, are how injected text steals data.

The rule's general advice: Do not install unless you can explain why this text needs those files. (see the review note for this case)

Known false alarm

The full sentence is 'Before running any script, first cd to that directory or use the full path', an ordinary usage note.

Reviewed by Claude (AI-assisted review), 2026-11-21

3 low-severity notes (informational, do not affect the result)

credential-access: Touches credentials or secrets low

skills/huggingface-paper-publisher/scripts/paper_manager.py:29

OWASP LLM LLM03 Supply ChainOWASP Agentic ASI03 Identity and Privilege AbuseOWASP Agentic ASI04 Agentic Supply Chain VulnerabilitiesMITRE ATLAS AML.T0055 Unsecured CredentialsMITRE ATLAS AML.T0083 Credentials from AI Agent Configuration

from dotenv import load_dotenv

What this means: Reads files or environment holding keys, tokens or sessions.

What you can do: Confirm the tool needs this; otherwise do not install.

fetch.auth-required: Remote server requires sign-in low

[server huggingface-skills] remote server answered HTTP 401; sign-in required

What this means: The remote server requires sign-in; its tool descriptions could not be read. Bawwab re-checks at session start.

What you can do: Sign in through Claude; Bawwab re-checks the tool descriptions at session start.

Dispute this result

Maintainer and think this is wrong? Open a dispute on GitHub with the target id plugin-huggingface-skills-78399c.