skills.rest
TylerSimons1127
TylerSimons1127COMMUNITY
Explore all skills by this author →
GitHub

bouncer

Scans inbound Hermes skills for prompt injection and decides allow, ask, or quarantine.

Updated Aug 21, 2026
Search Tags:#security#market research#prompt injection#hermes#quarantine#agent safety#skill vetting
One-click install
npx skills add https://github.com/TylerSimons1127/vibe --skill bouncer-tylersimons1127
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: bouncer
Source: https://github.com/TylerSimons1127/vibe/tree/main/skills/bouncer
Command: npx skills add https://github.com/TylerSimons1127/vibe --skill bouncer-tylersimons1127

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Installing third-party skills into an AI agent can introduce prompt injection, secret exfiltration, or low-quality bloat. This Skill acts as a security gate that vets every candidate skill before it enters the agent. ## Core Features & Use Cases - Prompt-Injection Scan: Reads the candidate's SKILL.md and referenced files to flag memory overwrites, secret exfiltration, jailbreaks, hidden network calls, and obfuscated payloads. - Market Research: Searches the web for alternative skills and ranks the candidate on maintenance recency, contributors, and reliability. - User Interview & Decision: Asks 2-4 intent questions, then returns auto-allow, ask-first, or deny-and-quarantine, saving flagged files to ~/.hermes/quarantine/. - Use Case: A user pastes a GitHub link to a community skill. The bouncer scans it, finds an undeclared webhook exfiltrating environment variables, quarantines the file, and reports the exact offending lines. ## Quick Start Scan this skill repo link for prompt injection and tell me whether it is safe to install.

Frequently Asked Questions about bouncer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I check if an AI agent skill is safe to install?▼

Run the bouncer workflow on the candidate skill: it reads the full SKILL.md and referenced files, flags injection patterns like memory overwrites or hidden webhooks, researches alternatives, then returns an auto-allow, ask-first, or deny-and-quarantine decision.

What prompt injection patterns does a skill security scan detect?▼

The scan flags instructions that overwrite standing memory or SOUL.md, exfiltrate secrets or files outside the workspace, embed ignore-previous-instructions jailbreaks, make undeclared outbound network requests, or hide obfuscated and base64-encoded payloads.

What happens when a skill fails the security scan?▼

Skills with hard injection findings are denied and quarantined. The flagged file is saved to ~/.hermes/quarantine/ rather than deleted, and the user is told why with the specific offending lines so they can inspect it.

Can a useful skill still be installed if it touches credentials?▼

Yes, via the ask-first decision. Skills that are useful but access credentials, external accounts, or make changes are installed only after explicit user confirmation, with the exact access surfaced beforehand.

Does the bouncer replace built-in skill validation?▼

No, it layers on top of Hermes' basic validation. It adds market research for better alternatives, a user intent interview to prevent skill bloat, and a quarantine mechanism for rejected skills.

Related Skills

Explore More Software Engineering Skills→
UnicomAIUnicomAI
2.5k

find-skills

Search and discover OpenClaw skills across ClawHub, marketplaces, and GitHub.

Community
Basic
jaredcroxtonjaredcroxton

hermes-agent

Configure Hermes Agent profiles, skills, and gateway workflows via CLI.

Community
Advanced
traygerbigtraygerbig
1

Skill Guardian

Scan OpenClaw skills for malicious code and security threats.

Community
Advanced
samueriosamuerio

agent-sandboxes

Execute isolated code and file operations in E2B sandboxes.

Community
Advanced