Index of every page on this site: llms.txt
All skills
openclaw avatar

/report-clawhub-malicious-skill

@7b51dc8
by openclawopenclaw/clawscan156 stars
30
Run it yourself

Use when a researcher, maintainer, or contributor found or suspects a malicious skill on ClawHub and needs a private reporting workflow: opening a GitHub private vulnerability report, preparing a proposal-only PR, creating `proposals/<GHSA-ID>/clawscan.yml`, running ClawScan against the malicious skill, and explaining what evidence belongs in private versus public channels.

Use this Skill: https://skilld.dev/gh/openclaw/clawscan/report-clawhub-malicious-skill

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈102 tokens for metadata: the name and description. ≈1k when used: this file. ≈67 more on demand in 1 file.

Description uses 5.1% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈409 characters, or 5.1%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Report ClawHub Malicious Skill

Overview

Use this workflow to help someone report a malicious ClawHub skill and propose a ClawScan profile change that catches it. Keep sensitive details private and make the public PR contain only the candidate ClawScan config.

Safety Boundary

Separate private evidence from public contribution material:

  • Put live malicious skill URLs/slugs, impact, exploit details, reproduction notes, and private ClawScan artifacts in GitHub private vulnerability reporting.
  • Do not paste secrets, exploit payloads, private artifacts, or live bypass details into public issues, public PR text, README edits, or config comments.
  • Do not ask the reporter to execute the suspicious skill. ClawScan should scan a local copy as data; it should not run the skill's behavior.
  • If the reporter does not have a GHSA/private report id yet, have them open the private report first and wait for the identifier before opening the public proposal PR.

Workflow

  1. Open a GitHub private vulnerability report.

    Ask the reporter to include:

    • affected ClawHub skill URLs or slugs
    • why the current clawhub profile missed it
    • why the proposed ClawScan config catches it
    • local ClawScan artifact paths or attached artifacts
    • any private reproduction context maintainers need
  2. Create a proposal-only branch and config.

    The public PR should add only:

    proposals/<GHSA-ID>/clawscan.yml

    The file must define a clawhub profile. Start from the current profile and change only what is needed to catch the reported case:

    version: 1
    profiles:
      clawhub:
       scanners:
         - skillspector
         - aig
        judge:
          command: >
            # candidate judge command, if changed

    Do not edit the official bundled profile files in the proposal PR:

    • internal/profiles/clawhub/clawscan.yml
    • internal/profiles/clawhub/prompt.md
    • internal/profiles/clawhub/output.schema.json
  3. Prove the candidate catches the reported skill locally.

    Use a local copy of the suspicious skill:

    clawscan /path/to/suspect-skill \
      --config proposals/<GHSA-ID>/clawscan.yml \
      --profile clawhub \
      --output ./artifacts/reported-skill-candidate.json

    If useful, compare against the built-in profile:

    clawscan /path/to/suspect-skill \
      --profile clawhub \
      --output ./artifacts/reported-skill-current.json

    The public PR should not include the private artifacts unless maintainers say the contents are safe to publish. Reference them in the private report.

  4. Open the public PR.

    Keep the PR text minimal:

    • state that a private vulnerability report exists
    • point to proposals/<GHSA-ID>/clawscan.yml
    • do not include sensitive skill details
    • say maintainers should run the official SkillTrustBench Profile Gate
  5. Explain maintainer validation.

    Maintainers run the official gate against the proposal:

    clawscan benchmark SkillTrustBench \
      --ids https://huggingface.co/datasets/cuhk-zhuque/SkillTrustBench-results/resolve/main/data/evaluation_subset_10pct.jsonl \
      --config proposals/<GHSA-ID>/clawscan.yml \
      --profile clawhub \
      --output ./artifacts/skilltrustbench-candidate.json

    The proposal's clawhub profile shadows the built-in profile for that run. Maintainers review the private report, upload/preserve the benchmark artifact, add a dated baseline under benchmarks/skilltrustbench-leaderboard-10pct/, and port accepted behavior into the bundled clawhub profile. The latest YYYY-MM-DD.json baseline is the current accepted baseline.

Output Shape

When walking someone through the process, end with:

  • private report checklist
  • exact proposal file path
  • candidate config draft or edits needed
  • local ClawScan commands to run
  • public PR checklist
  • clear note that maintainers own the official SkillTrustBench gate and final built-in profile port

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts2mo3 checks · Risk SAFE
  • Gen Agent Trust Hub2mo

    The skill establishes a secure workflow for reporting malicious skills and proposing improvements to analysis profiles. It uses trusted external sources and emphasizes privacy through official vulnerability reporting channels.

  • Socket2mo

    No alerts

  • Snyk2mo

    Risk: LOW · No issues

Signed by skilld at 7b51dc8. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated 2 weeks ago

README badge

README badge for openclaw/clawscan/report-clawhub-malicious-skill