More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    RepoTrap: Prompt-Injection & Repo Trap Scanner

    1

    Scan cloned repos, skills, GitHub issues & transcripts for invisible Unicode and prompt-injection traps

    $6

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    RepoTrap: Prompt-Injection & Repo Trap Scanner

    RepoTrap: Prompt-Injection & Repo Trap Scanner

    Example session with this skill installed

    Scan my installed skills, local repos, and Claude Code / Codex transcripts with repotrap and check for invisible Unicode steganography or repo-open execution traps.

    • Read your context and instructions
    • Compiled the repotrap

    RepoTrap Audit Summary

    Scanned 525 installed skills, 16 workspace repos, and 153 Claude Code & Codex CLI sessions (Safety Score: 64/100, Grade C).

    MetricRaw RegexRepoTrap Filtered
    <system-reminder> hits1,5260 untrusted (1,526 native attachment rows filtered via RTR012)
    Zero-width chars in tool outputs4142 across 15 outputs in 5 sessions (including Codex nested object outputs via RTR011)
    Skill remote shell pipes (`curlsh`)1

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

    About this skill

    Running coding agents on untrusted repositories or third-party issues exposes your environment to indirect prompt injection and malicious configuration traps. This skill audits local files, GitHub PRs, and session history to neutralize these threats before execution.

    What it does

    • Repo-Open Trap Detection flags malicious configurations in .claude/settings.json, .mcp.json, and .vscode/tasks.json that execute code on folder open.
    • Steganography Scanning identifies invisible Unicode tags, Bidi overrides, and zero-width spaces used to hide instructions from humans while hijacking LLMs.
    • Delimiter Spoofing Audit detects fake <system-reminder>, </tool_result>, or hidden HTML comments designed to bypass agent guardrails.
    • Transcript Analysis parses .jsonl session history from Claude Code or Codex to find evidence of past injection attempts.
    • Automated Sanitization generates a PostToolUse hook to strip malicious patterns from tool outputs in real-time.

    How it works

    1. Audit Workspace by running the scanner against local repositories or installed SKILL.md packages to find dormant configuration traps.
    2. Scan External Input using the --gh-issue flag to inspect untrusted GitHub issues or PR bodies before allowing an agent to ingest them.
    3. Install Hook by emitting the deterministic sanitizer script and merging it into your agent settings to provide persistent protection.

    Frameworks & tools

    The skill uses a zero-dependency Node.js scanner compatible with Claude Code, Cursor, and Codex CLI environments. It integrates with the GitHub CLI for remote issue auditing.

    Why this beats prompting it yourself

    LLMs are inherently susceptible to the very injections they are trying to detect, especially invisible Unicode. This skill uses deterministic regex and hex-value scanning that cannot be bypassed by "ignore previous instructions" prompts.

    Use cases

    • Auditing a cloned repository before opening it in an agent-enabled IDE.
    • Pre-scanning GitHub issues from unknown users before using gh issue view.
    • Cleaning session transcripts to prevent poisoning of long-term agent memory.
    • Enforcing safety standards across a team of developers using coding assistants.

    Known limitations

    Requires Node.js for scanner execution. GitHub issue scanning requires the GitHub CLI (gh) to be authenticated and installed.

    Requires Node.js environment. GitHub issue scanning requires authenticated GitHub CLI (gh). Scanner is deterministic and may flag legitimate Unicode in non-English documentation as false positives.

    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 11 days ago

    Requires

    Node.js

    Permissions

    Terminal / Shell
    Network Access
    Listed11 days ago

    What's inside

    Frequently Asked Questions

    Standard LLMs often fail to 'see' hidden Unicode or zero-width characters that can hijack their logic. Repotrap uses a deterministic engine to scan hex-values directly, ensuring it catches injections that would bypass a standard prompt-based guardrail.

    Repotrap is optimized for CLI-based agents like Claude Code and Codex, as well as IDEs like Cursor. It includes a specific feature to merge sanitization hooks into your agent's settings for persistent protection across sessions.

    The 'Repo-Open Trap' detection specifically scans configuration files like `.vscode/tasks.json` or `.mcp.json` for auto-run commands. These traps can execute malicious scripts the moment a developer—or an AI agent—simply opens a project folder.

    Yes, using the GitHub CLI integration, you can scan the body and comments of remote PRs or issues before your agent ingests them, preventing 'indirect prompt injection' from external contributors.

    You receive the full deterministic scanner engine, the automated sanitization hook script, and integration guides for major agent environments. No external API keys are required for the scanner itself.

    New in Security