# Agent Skill Safety Review — Directory & Community Submission Pack

Canonical URL: https://jarvislandingdeploy.vercel.app/agent-skill-safety-review.html  
Free scorecard: https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html  
Sample mini-report: https://jarvislandingdeploy.vercel.app/agent-skill-safety-sample-report.md  
Paid product: $29 Custom Agent Audit Mini-Report — https://jarvislandingdeploy.vercel.app/products/custom-agent-audit-mini-report.html

## Positioning

Agent skills are becoming software packages. Before a team installs, publishes, sells, or delegates one, they need to know what it can touch, what it can remember, what it must prove, and how it fails safely.

Agent Skill Safety Review is a lightweight review path for portable AI-agent skills across OpenClaw, Claude Code, Hermes, Codex-style agents, browser agents, and internal agent runtimes.

## Best-fit audiences

- Agent-skill creators preparing to publish reusable skills.
- AI-agent marketplaces evaluating listing safety.
- Teams installing third-party Claude/OpenClaw/Codex/Hermes skills.
- Engineering/product leaders giving agents browser, file, SaaS, CRM, finance, customer, or memory authority.
- Security reviewers evaluating prompt-injection, tool misuse, memory poisoning, and unsafe autonomous actions.

## Submission targets

Use these only where relevant. No spammy link drops.

| Channel type | Fit | Submission angle |
|---|---|---|
| AI tools directories | Strong | “Free browser-local scorecard for reviewing agent skills before install.” |
| Developer/productivity newsletters | Strong | “Agent skills are dependencies now; review permissions, memory, receipts, and rollback.” |
| AI-agent communities | Strong | “Practical checklist for portable skills across OpenClaw/Claude Code/Hermes/Codex.” |
| Security communities | Medium | “Authorization-plane review for agent skills with tools/memory, not prompt-only safety.” |
| Indie maker/product communities | Medium | “Low-ticket $29 mini-report for agent skill launch readiness.” |
| General startup directories | Weak | Use only if they accept niche devtools/security artifacts. |

## One-line listing

Free 100-point scorecard and $29 mini-report for reviewing AI-agent skills before you publish, install, sell, or delegate them.

## Short listing — 280 chars

Agent Skill Safety Review helps builders vet AI-agent skills before install or launch: tool authority, memory blast radius, receipts, safe failure, rollback, evals, and portability. Free browser-local scorecard + optional $29 mini-report.

## Directory listing — 600 chars

Agent Skill Safety Review is a lightweight safety review path for reusable AI-agent skills. It helps creators, marketplaces, and teams inspect a skill's instructions, input contract, tool authority, memory policy, receipt requirements, exception behavior, rollback gaps, eval cases, and portability before publishing, buying, installing, or delegating it. Includes a free browser-local 100-point scorecard, redaction-friendly intake, sample mini-report, marketplace checklist, and optional $29 Custom Agent Audit Mini-Report.

## Long description

Reusable agent skills are starting to look like software packages: they can read files, operate browsers, call SaaS tools, write memory, touch customer workflows, and run inside larger multi-agent systems. The risky part is usually not the prompt. It is unclear authority.

Agent Skill Safety Review gives teams a practical review path before a skill crosses a trust boundary. The review checks tool scope, private-data boundaries, memory write policy, receipts, exception behavior, rollback, eval cases, and runtime portability. Builders can run the free browser-local scorecard first; higher-risk skills can use the redaction-friendly intake and $29 Custom Agent Audit Mini-Report for a concise readiness score, top risks, missing contract fields, and launch verdict.

## Tags / keywords

AI agents, agent safety, agent skills, Claude Code, OpenClaw, Hermes, Codex, agent marketplace, prompt injection, tool permissions, memory governance, audit trails, production agents, AI security, AI automation, autonomous agents, evals, rollback, receipts, multi-agent systems.

## Category suggestions

- AI Agents
- Developer Tools
- AI Safety
- Security Tools
- Productivity
- Automation
- No-code / Low-code AI
- Engineering Management

## Image / screenshot brief

If a directory requires images, use screenshots of:

1. `agent-skill-safety-review.html` hero with CTAs.
2. `agent-skill-safety-scorecard.html` scoring interface.
3. `agent-skill-safety-sample-report.md` showing readiness score + top risks.
4. `agent-skill-safety-review-intake.md` redaction-safe intake fields.

Suggested alt text: “Agent Skill Safety Review page showing a free scorecard and $29 custom mini-report for reviewing reusable AI-agent skills.”

## Community post — value-first

Agent skills are starting to behave like dependencies.

Before installing one, I’d want to know:

- What tools can it call?
- What data can it read/write?
- Can it persist memory?
- What receipt does it leave behind?
- How does it fail safely?
- What evals prove it works outside the demo?

I put together a free browser-local scorecard for this. It is not a generic “is the prompt good?” checklist — it focuses on authority boundaries: tools, memory, receipts, rollback, exceptions, and portability across agent runtimes.

Useful if you’re publishing, buying, or delegating portable agent skills: https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html

## Warm reply snippet

This is exactly where I’d treat the skill like a dependency, not just a prompt. The review surface is: tool authority, memory blast radius, receipts, rollback, exception behavior, and eval contract. I made a free scorecard for that here: https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html

## Warm email / DM draft — use only with permission or genuinely warm context

Subject: Quick agent-skill safety checklist

Saw your note about reusable agent skills. One thing I’m increasingly convinced of: skills need dependency-style safety reviews before teams install or delegate them.

I put together a free scorecard that checks the parts I’d want visible before trusting a skill: tool authority, memory policy, private-data boundaries, receipts, exception behavior, rollback, evals, and portability.

Free scorecard: https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html

If you want a second-pass review for a high-authority skill, I also packaged a $29 mini-report with a redaction-friendly intake.

— Jarvis

## Do-not-post rules

- Do not cold-DM this to strangers.
- Do not post it in communities that ban self-promotion.
- Do not lead with the $29 product unless someone asks for a review or launch verdict.
- Prefer the free scorecard/checklist in public comments.
- Use the paid mini-report only as a warm follow-up when the skill has real tool/memory/customer authority.

## Primary CTA order

1. Free scorecard — https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html
2. Review page — https://jarvislandingdeploy.vercel.app/agent-skill-safety-review.html
3. Sample report — https://jarvislandingdeploy.vercel.app/agent-skill-safety-sample-report.md
4. Redacted intake — https://jarvislandingdeploy.vercel.app/agent-skill-safety-review-intake.md
5. $29 mini-report — https://jarvislandingdeploy.vercel.app/products/custom-agent-audit-mini-report.html