More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    SkillJail Pro — 10 Tested Security Profiles

    1

    Ten hardened, kernel-verified least-privilege profiles for common skills.

    $5.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    SkillJail Pro — 10 Tested Security Profiles

    SkillJail Pro — 10 Tested Security Profiles

    Example session with this skill installed

    Lock down my Postgres reporting skill with least privilege so it can run its queries but can't exfiltrate the database if it's ever prompt-injected.

    • Read your context and instructions
    • Compiled the skilljail pro

    Summary

    Drop in the postgres-query profile. The skill can reach only your database host; a dump to anywhere else is blocked, and it can't read credentials on disk.

    The profile (skilljail.yaml):

    permissions:
      fs:   { read: ["$WORKSPACE"], write: ["$WORKSPACE/out"] }
      net:  { allow: ["*.rds.amazonaws.com", "*.supabase.co", "*.neon.tech"] }
      exec: { allow: ["psql", "python3", "cat", "grep", "awk", "sort", "cut"], shell: true }
      env:  { pass: ["DATABASE_URL", "PGPASSWORD"] }
    

    Verified under the kernel (scripts/verify.sh):

    === postgres-query ===
      [ok] manifest valid
      [ok] legit workflow ran clean
      [ok] attack A blocked: ~/.aws/credentials unreadable
      [ok] attack B blocked: disallowed binary (nc) cannot exec
    ALL PROFILES PASS
    

    Exfiltration attempt to an un-allowed host (denied):

    curl: (56) CONNECT tunnel failed, response 403
    [skilljail] net deny: CONNECT attacker-exfil.example.com:443 — no net.allow rule matches
    

    Next steps

    • Edit net.allow to your real DB host and you're done; repeat for the other 9 skills.

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

    About this skill

    SkillJail (free) jails any skill to exactly what it declares, enforced at the kernel. The fiddly part is writing a manifest that is tight enough to stop an attack but loose enough that the real skill still works — too loose and it is useless, too tight and the skill breaks on a missing cat or a git re-exec. This pack does that work for you, for the ten skill types people actually run, and proves each one under the kernel.

    Profiles included: git-commit · github-pr · postgres-query · s3-sync · api-request · slack-webhook · web-scrape · csv-transform · json-transform · npm-build

    Each ships a ready manifest (skilljail.yaml), an example SKILL.md, and a legit.sh you can run under the jail to confirm it works on your machine. Every one is verified: the real workflow runs clean, reading ~/.aws/credentials is denied, a disallowed network binary is denied, and network-restricted profiles block egress to any host outside their allowlist.

    Also included:

    • HARDENING.md — how to write and tune your own tight manifests (the macOS git-shim gotcha, the exec-allowlist rules, CI patterns)
    • catalog.md — each profile's protection summary and verification result
    • scripts/verify.sh — re-run the legit-plus-attacks check on any profile, on your machine

    Requires SkillJail (free, Apache-2.0). Commercial license; redistribution of the pack is not permitted.

    Requires SkillJail (free, Apache-2.0) installed on the host. Users must manually define specific allowed hosts and bucket names in the YAML manifests.

    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

    Permissions

    Terminal / Shell
    Write Files
    Network Access

    Allowed Hosts

    api.example.com
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    These profiles protect against 'agent escape' scenarios where a compromised or hallucinating agent tries to steal SSH keys, access internal network metadata, or exfiltrate environment variables. Unlike LLM prompting, these restrictions are enforced at the system kernel level, making them impossible for an agent to bypass via prompt injection.

    The purchase includes ten production-ready YAML manifests, a suite of 'Attack Verification' scripts to test your defenses, and detailed hardening documentation for complex environments like macOS and npm. You also receive the configuration hooks needed to integrate these profiles with the SkillJail CLI.

    Yes, while the profiles are pre-configured for common tasks like Git or S3 access, you are expected to edit the 'net.allow' section to specify your exact database endpoints or API hosts. This ensures the 'least-privilege' model is tailored specifically to your infrastructure.

    These profiles are compatible with any AI agent or framework as long as the host machine has the SkillJail open-source library installed. They work by wrapping the agent's execution environment, meaning they are agnostic to the specific LLM (like GPT-4 or Claude) being used.

    The profiles are designed to allow standard operations like git re-execs and npm builds that often trigger 'false positives' in stricter security setups. The included documentation provides specific allow-lists for these edge cases to ensure your agent stays productive while remaining locked down.

    New in Security