More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    🛡️ Skill Injection Scanner

    1

    Review Markdown and YAML skill files for suspicious instruction patterns, with source excerpts withheld and incomplete inputs clearly flagged. Requires Python 3.12+.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    🛡️ Skill Injection Scanner

    🛡️ Skill Injection Scanner

    Example session with this skill installed

    Run python scripts/scan_skill_injection.py fixtures/risky-sample on the included synthetic fixture. Report the detector output without executing anything in the fixture.

    • Read your context and instructions
    • Compiled the injection scanner

    Skill Injection Scanner Report

    Status: REVIEW (not a safety certification)

    Files scanned: 2
    Findings: 9

    [CRITICAL] SIS001 - fixtures\risky-sample\SAMPLE-SKILL.md:15

    Secret or environment reference is paired with outbound wording.
    Evidence: [withheld: inspect the cited source locally]

    [HIGH] SIS002 - fixtures\risky-sample\SAMPLE-SKILL.md:11

    Conditional trigger includes a side effect.
    Evidence: [withheld: inspect the cited source locally]

    [HIGH] SIS004 - fixtures\risky-sample\SAMPLE-SKILL.md:15

    HTML comment contains imperative instruction wording.
    Evidence: [withheld: inspect the cited source locally]

    [MEDIUM] SIS005 - fixtures\risky-sample\SAMPLE-SKILL.md:17

    Long base64-looking blob found.
    Evidence: [withheld: inspect the cited source locally]

    [MEDIUM] SIS006 - fixtures\risky-sample\SAMPLE-SKILL.md:17

    Single unusually long token found.
    Evidence: [withheld: inspect the cited source locally]

    [HIGH] SIS007 - fixtures\risky-sample\SAMPLE-SKILL.md:13

    Remote content is described as instructions to follow.
    Evidence: [withheld: inspect the cited source locally]

    [HIGH] SIS008 - fixtures\risky-sample\SAMPLE-SKILL.md:19

    Unsafe command behavior is described as an instruction risk.
    Evidence: [withheld: inspect the cited source locally]

    [HIGH] SIS008 - fixtures\risky-sample\SAMPLE-SKILL.md:19

    Unsafe command behavior is described as an instruction risk.
    Evidence: [withheld: inspect the cited source locally]

    [MEDIUM] SIS009 - fixtures\risky-sample\agents\openai.yaml:1

    Broad tool access is requested without a scope justification.
    Evidence: [withheld: inspect the cited source locally]

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Review a skill before you enable it

    Skill Injection Scanner applies local text heuristics to Markdown and YAML files in a stable skill-package copy. It flags secret-and-outbound wording, conditional side effects, hidden comments, invisible characters, encoded-looking blobs, remote instruction fetches, unsafe-command references and broad capability requests.

    What the report tells you

    Each finding retains its rule, severity, file and line. Version 1.1.0 withholds source excerpts instead of repeating potentially sensitive text. CANNOT_ASSESS means inputs were incomplete; REVIEW means findings need interpretation; NO_MATCHES means only that no rule matched within the inspected scope.

    Requirements and limits

    Requires local Python 3.12+ and permission to read the selected package. Standard library only; no network or package edits. Missing, unreadable, invalid-UTF-8 and linked inputs are reported rather than silently treated as clean. The detector does not scan Python or shell source, execute the package, test runtime defenses, certify safety or protect against concurrent filesystem changes. Paths may still contain private information. Review findings manually and inspect code and provenance separately.

    Heuristic Markdown/YAML screening only. Requires local Python 3.12+ and a stable package copy. Does not inspect Python or shell source, execute packages, certify safety or protect against concurrent filesystem changes. No matches does not mean safe; CANNOT_ASSESS means inputs were incomplete. Source excerpts are withheld, but file paths may be sensitive. False positives and missed attacks are possible.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Python 3.12+

    Permissions

    Terminal / Shell
    Read Files

    File Scopes

    skill-injection-scanner/**
    **/*.md
    **/*.yaml
    **/*.yml

    Read-only. Applies standard-library text heuristics and prints findings with rule id, severity, file, and line. It never executes the content it scans and reads no environment variables.

    Listed5 months ago
    Updated1 month ago

    What's inside

    Frequently Asked Questions

    No. This is a heuristic Markdown/YAML screening pass. Findings require manual interpretation; no matches is not a safety certificate.

    It looks for suspicious indicators like environment-variable-to-URL exfiltration, obfuscated base64 blobs, zero-width characters, and overbroad tool requests like simultaneous network and filesystem access.

    A SKILL.md-capable agent must be able to run local Python 3.12+. The bundled detector reads Markdown and YAML, not Python or shell source. Individual agent-host compatibility is not certified.

    No. Review a stable local copy of the package, not production secrets or logs. Reports withhold source excerpts, but file paths can still contain sensitive information.

    The package includes SKILL.md, a standard-library Python detector, a manual-review checklist, remediation guidance, synthetic clean/risky fixtures and regression tests.

    No. It runs offline using the patterns in the downloaded version. Review the changelog and exact package when an update is released. It can miss attacks outside those heuristics.

    New in Security