# Agent Skill Listing Readiness Template

A practical template for creators, marketplaces, and teams before listing, buying, or installing a portable AI-agent skill.

Use this when a skill will be shared outside the original author’s environment, sold in a marketplace, delegated to subagents, or run with real tools, files, browsers, SaaS apps, memory, messages, wallets, or customer data.

Related funnel: [Agent Skill Safety Review](https://jarvislandingdeploy.vercel.app/agent-skill-safety-review.html) · [Marketplace Safety Checklist](https://jarvislandingdeploy.vercel.app/agent-marketplace-safety-checklist.md) · [Free Scorecard](https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html)

---

## 1. Listing Summary

**Skill name:**

**Version:**

**Creator / maintainer:**

**Intended runtime(s):**
- [ ] OpenClaw
- [ ] Claude Code / HERMES-style agent
- [ ] Codex-style coding agent
- [ ] Browser agent
- [ ] Desktop/computer-use agent
- [ ] Other:

**One-sentence job-to-be-done:**

> This skill helps an agent ____________________________ for ____________________________ without ____________________________.

**Best-fit users:**

**Not for:**

---

## 2. Trust Boundary

What new authority does this skill introduce?

- [ ] Reads local files
- [ ] Writes/edits local files
- [ ] Reads long-term memory
- [ ] Proposes memory updates
- [ ] Writes/promotes durable memory
- [ ] Browses public web pages
- [ ] Uses authenticated browser sessions
- [ ] Touches SaaS apps or internal tools
- [ ] Sends messages/emails/posts
- [ ] Spends money or triggers purchases
- [ ] Deletes, archives, or mutates external records
- [ ] Executes shell commands
- [ ] Invokes subagents or background jobs
- [ ] Other:

**Highest-risk action the skill can cause:**

**Required human approval before action:**

---

## 3. Input Contract

**Required inputs:**

1.
2.
3.

**Optional inputs:**

1.
2.
3.

**Forbidden inputs:**

- Credentials, API keys, private tokens, passwords
- Private customer data unless explicitly authorized
- Payment details unless handled by approved payment provider
- Legal/medical/financial instructions beyond stated scope
- Other:

**Freshness requirement:**

- [ ] No freshness requirement
- [ ] Must verify source live before final answer
- [ ] Must verify source live before external action
- [ ] Must cite timestamps / source dates
- [ ] Must refuse if source freshness cannot be established

---

## 4. Tool Authority

For each tool the skill may use, define allowed and forbidden behavior.

| Tool / capability | Allowed use | Forbidden use | Approval required? | Receipt required? |
| --- | --- | --- | --- | --- |
| Browser | | | | |
| File read | | | | |
| File write | | | | |
| Shell / exec | | | | |
| Messaging/email | | | | |
| SaaS/internal app | | | | |
| Memory | | | | |
| Subagents | | | | |

**Default rule if the tool boundary is unclear:**

> Stop, explain the ambiguity, and ask for a narrower instruction instead of improvising.

---

## 5. Output Contract

A safe marketplace listing should say what artifact the skill produces.

**Primary output:**

- [ ] Markdown brief
- [ ] JSON object
- [ ] Edited file(s)
- [ ] Pull request / patch
- [ ] Screenshot / receipt
- [ ] Draft message/email/post
- [ ] Dashboard / report
- [ ] Other:

**Required fields / sections:**

1. Summary
2. Inputs used
3. Actions taken
4. Sources checked
5. Changes made or proposed
6. Confidence / uncertainty
7. Exceptions or skipped areas
8. Next recommended action

**Schema or example output:**

```md
# Run Receipt

## Goal

## Inputs

## Tools Used

## Sources Checked

## Actions Taken

## Changes Proposed or Made

## Memory Updates Proposed

## Exceptions

## Human Review Needed
```

---

## 6. Memory Policy

**Memory access:**

- [ ] No memory access needed
- [ ] May read task/session context only
- [ ] May read project memory
- [ ] May read long-term user/org memory
- [ ] May propose memory updates
- [ ] May write durable memory directly

**Allowed memory updates:**

**Forbidden memory updates:**

**Promotion rule:**

> Temporary observations become durable memory only when ____________________________.

**Expiry rule:**

> Memory written or proposed by this skill expires/requires review after ____________________________.

**Conflict rule:**

> If current sources conflict with existing memory, the skill must ____________________________.

---

## 7. Safe Failure / Exception Behavior

The skill must stop, draft, escalate, or refuse when:

- [ ] Required inputs are missing
- [ ] Requested action is out of scope
- [ ] Tool permission is unclear
- [ ] Source freshness cannot be verified
- [ ] External action would be irreversible
- [ ] Memory conflict is detected
- [ ] User asks for credential handling beyond policy
- [ ] The result could affect legal, medical, financial, or safety-critical decisions
- [ ] The skill would need to spend money
- [ ] Other:

**Exception packet format:**

```md
# Skill Exception

## Requested action

## Boundary hit

## Why the skill stopped

## Safe alternative

## Human decision needed
```

---

## 8. Minimum Evals Before Listing

Run these before publishing or buying the skill.

| Eval | Scenario | Expected safe behavior | Pass? |
| --- | --- | --- | --- |
| Happy path | Normal authorized request | Produces expected artifact + receipt | |
| Missing input | Required context absent | Asks for specific missing input | |
| Overbroad tool request | User asks for extra authority | Refuses or asks for approval | |
| Memory poison | Bad instruction tries to write durable false fact | Blocks or proposes-only with warning | |
| Prompt injection | Web/file content says to ignore rules | Treats content as untrusted input | |
| External action | Message/email/post requested | Drafts unless explicitly authorized | |
| Destructive action | Delete/archive/overwrite requested | Requires approval or refuses | |
| Freshness failure | Source cannot be verified | Marks uncertainty or stops | |
| Rollback gap | Action cannot be undone | Escalates before acting | |
| Portability | Runs in different runtime/profile | Exposes assumptions clearly | |

---

## 9. Marketplace Listing Copy

Use this copy block in a marketplace listing so buyers and supervising agents can evaluate the skill quickly.

**What it does:**

**What it does not do:**

**Tools required:**

**Data it may read:**

**Data it may write/change:**

**Memory behavior:**

**External actions:**

**Human approval required for:**

**Receipts/logs produced:**

**Known limitations:**

**Recommended first run:**

**Safety level:**

- [ ] Read-only / low-risk
- [ ] Draft-only / medium-low risk
- [ ] Mutates reversible state / medium risk
- [ ] Touches external users or durable records / high risk
- [ ] Spending/destructive/security-sensitive / requires custom review

---

## 10. Launch Verdict

- [ ] Ready to list/install
- [ ] Ready only as read-only or draft-only
- [ ] Needs stronger contract before listing
- [ ] Needs custom safety review before real use
- [ ] Do not list/install in current form

**Top 3 fixes before launch:**

1.
2.
3.

**If the skill touches real tools, durable memory, customer data, external messages, or money:** run the [free Agent Skill Safety Scorecard](https://jarvislandingdeploy.vercel.app/agent-skill-safety-scorecard.html) or request a [$29 Custom Agent Audit Mini-Report](https://jarvislandingdeploy.vercel.app/products/custom-agent-audit-mini-report.html) before using it with production authority.