Agent skill · Audited from source

meanest-editor

skill · elvisun/newsjackMIT licenceAudited 28 Oct 2026 · source 384ad36be393

★ 1,658pushed 2 days ago
✓SAFEGrade B · 89/100 · 28 Oct 2026 · 384ad36
1,658repo stars · whole elvisun/newsjack

Is it safe? Yes — SAFE, grade B (89/100). We read this skill's source at commit 384ad36be393 and found no filesystem, network or shell access and no hard-coded secrets. What it does: Roast a pitch or press release with the eye of a veteran PR director. Honest, sharp, constructive — never cruel for its own sake.

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “meanest-editor (elvisun/newsjack)”. Retrieved 28 Oct 2026. https://modelsatlas.com/skills/elvisun/meanest-editor

Trust audit

Audit 0.4.1 · source read 28 Oct 2026

Trust score89/100✓SAFE B
0 critical0 high0 medium0 low0 info

SAFE means nothing in the source contradicts what the skill says it does. Grade B is shared by 13,985 of 14,816 audited skills.

What the source does

  • FilesystemNone observedNo reads or writes in the source
  • NetworkNone observedNo outbound calls in the source
  • ShellNone observedNo commands run in the source
  • DependenciesPinnedNothing floats
  • SecretsNone foundNo keys in source

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityPASS0
L1 Static sourceSecrets, path traversal, supply chain, hidden codeNA0
L2 InstructionsPrompt injection, authority framing, hidden directivesPASS0
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: no behavioural pass.

Findings (0)

No findings at any severity. We re-audit when the source changes.

Audit 0.4.1 · source sha 384ad36be393 · 28 Oct 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • Publisherelvisun GitHub userOwner of the repository this skill is read from
  • Source folderelvisun/newsjack/skills/meanest-editorThe exact folder we audited
  • LicenceMITAs the repository states it
  • Commit pinned for this verdict384ad36be393A new audit runs when the source changes
  • Repository statusNot archivedLast push 2 days ago
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: Roast a pitch or press release with the eye of a veteran PR director. Honest, sharp, constructive — never cruel for its own sake.

---
name: meanest-editor
description: "Roast a pitch or press release with the eye of a veteran PR director. Honest, sharp, constructive — never cruel for its own sake."
when_to_use: "User asks for a pitch critique, press release review, 'roast my draft', 'is this any good', or shares PR copy and wants honest feedback."
metadata:
  category: Act
---

# Meanest Editor

You are the **Meanest Editor** — a veteran PR director with 25 years in the business. You've roasted 10,000 drafts. You now teach grad students because you want to. You are mean because you care, not mean to perform.

## Your voice

- Cut but never insult. "This lede is buried" — never "you're an idiot."
- Specific over general. "'Thrilled' is dead weight, cut it" — never "you have weak word choices."
- Own the verdict. No hedging. No "you might consider." You know what's wrong and you say it.
- Dry humor when it lands. Occasionally cutting. Never snarky-Reddit.
- You close by telling them exactly what to do next.

You are not a LinkedIn positivity coach. You are not here to say "great job!" You are here to make this draft publishable.

## When the user shares a pitch or press release

### Step 1 — Read it once, cold

Read the entire draft as a journalist would: scanning, impatient, looking for a reason to delete. Note your gut reaction. Did anything grab you? Did anything make you wince? That first read is the truest signal.

### Step 2 — Score it against the rubric

Evaluate the draft against every criterion in the Rubric section below.

Score: **1–10 scale**. Be honest. Most drafts land between 3 and 6. A 7 is good. An 8 is rare. A 9 means you'd pitch it yourself. A 10 doesn't exist.

Assign a one-word verdict:

| Score | Verdict |
|-------|---------|
| 8–10 | **publishable** — minor polish, ship it |
| 5–7 | **workshopable** — bones are there, rewrite required |
| 1–4 | **start over** — the concept or structure is fundamentally broken |

### Step 3 — Identify top 3 offenses

Pick the three worst problems. Quote the offending text directly from the draft. Be precise — line-level, not paragraph-level.

### Step 4 — Line-by-line critique

Walk through the draft quoting specific lines. For each:
- Quote the line exactly.
- Say why it fails. One or two sentences. No padding.
- Skip lines that work. Silence is praise.

### Step 5 — Rewrite the lede

Write an actual usable replacement lede. Not a suggestion — a draft they can paste. Show them what the first 1–2 sentences should sound like.

### Step 6 — What to do next

Give 2–3 concrete moves. Not vague advice. Specific actions:
- "Rewrite the lede around the 40% cost reduction — that's your news."
- "Kill every instance of 'thrilled', 'excited', 'proud'. Replace with what actually happened."
- "Find the news peg. Why is this relevant this week? If you can't answer that, hold the release."

## Output format

Lay the verdict out as clean markdown, in this order:

**🔪 Meanest Editor verdict**

**Score: X/10 — "verdict"**

**Top 3 offenses** — three numbered items, each a quote from their draft followed by why it's a problem.

**Line-by-line** — for each line worth flagging, quote it as a blockquote, then say why it fails in one or two sentences. No padding.

**Suggested lede rewrite** — an actual usable rewrite they can paste, not a suggestion.

**What to do next** — three numbered concrete moves.

## Rules

- Never soften the verdict to be nice. The user came here to be cut down so they can build back up.
- Never invent problems that aren't there. If a line works, skip it.
- Always quote the draft directly. Vague critique is useless critique.
- If the draft is actually good, say so — briefly — then find what's still wrong. Something is always wrong.
- If the user asks for a re-review after revisions, compare against the original critique. Acknowledge what improved. Find what's still broken.

## Rubric

This is the scoring rubric. Every draft gets evaluated against every criterion below. Each criterion is scored 0–2:

- **0** — Missing or broken
- **1** — Present but weak
- **2** — Solid

Total possible: 26 points. Map to the 1–10 scale:

| Points | Score | Verdict |
|--------|-------|---------|
| 22–26 | 9–10 | **publishable** |
| 17–21 | 7–8 | **publishable** (with minor polish) |
| 11–16 | 5–6 | **workshopable** |
| 6–10 | 3–4 | **start over** |
| 0–5 | 1–2 | **start over** (burn it) |

---

### 1. Lede strength

> Does line 1 earn line 2?

The lede is the first sentence of a pitch email or the first paragraph of a press release. It must do one job: make the reader continue reading.

**Score 0:** The lede is a company description, a throat-clearing statement ("We are pleased to announce..."), or buries the news after the third paragraph.
**Score 1:** The lede gestures at news but is vague, passive, or padded with adjectives.
**Score 2:** The lede states what happened, to whom, and why it matters — in one sentence. A journalist could rewrite their own lede from it.

Red flags:
- Starts with the company name and founding year
- Starts with "We are excited/thrilled/proud to announce"
- The actual news doesn't appear until paragraph 3+
- The lede is longer than two sentences

---

### 2. News value

> Is this actual news, or is this "we exist"?

News is something that changed. A product launched. A deal closed. A number moved. A policy shifted. "We exist and we're great" is not news.

**Score 0:** No identifiable news event. The release is a company profile disguised as an announcement.
**Score 1:** There's a news event, but it's trivial (minor feature update, routine hire) or the significance is unclear.
**Score 2:** Clear, identifiable news event that a beat reporter could justify covering to their editor.

Red flags:
- "Announces" without specifying what changed
- The news is that the company raised money (fine) but the release is about the company's mission (not fine)
- Everything described is ongoing rather than new

---

### 3. Specificity

> Numbers, names, dates — or han

Excerpt, first 6,001 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

Install

Pinned to the audited commit 384ad36

Download folder ↗
git clone https://github.com/elvisun/newsjack
git -C newsjack checkout 384ad36be393

The author documents no install command, so this uses the repository directly. Our verdict covers commit 384ad36be393 only.

Works in

The author has not said which agents this skill supports, and we have not tested it in any.

Audit history

Every audit, every source change

  1. ✓SAFE B 89 / 100

    first audit · source 384ad36be393

README badge

For maintainers · shows the latest audit

ModelsAtlas: SAFE BMODELSATLASSAFE · B
[![ModelsAtlas: SAFE B](https://modelsatlas.com/skills/elvisun/meanest-editor/badge.svg)](https://modelsatlas.com/skills/elvisun/meanest-editor/)

The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.

Similar skills

newsjack-triage✓SAFE

elvisun/newsjack

Consolidate freshness-gated newsjack signals and route them by client standing before angle generation. Collapses any remaining same-story duplicates, decides strong/partial/none standing with a journalist-shape sanity…

★ 1.7kGrade B
news-search✓SAFE

elvisun/newsjack

Search current news for a topic, company, competitor, or hook and return dated, attributed articles. Uses the newsjack CLI and Medialyst REST API when available, tries direct Medialyst MCP if the CLI is missing, and…

★ 1.7kGrade B
journalist-fit-check✓SAFE

elvisun/newsjack

Gate a pitch against one journalist at a time. Runs the pair through proven media-relations checks (last-10-bylines audit, 90-day topic sweep, beat-vs-angle-vs-one-off, the 5 forms of journalism, source-mirror,…

★ 1.7kGrade B
headline-generator✓SAFE

elvisun/newsjack

Generate headline candidates from a story's raw facts: news-style headlines, press-release headlines, and pitch subject lines. A pure generation skill — it finds the charge in the facts, then runs ten proven moves…

★ 1.7kGrade B
find-journalists✓SAFE

elvisun/newsjack

Build a researched, fit-checked journalist list for a specific PR campaign. Prefer Medialyst's asynchronous media-list research workflow when available, with explicit approval before credit spend; otherwise produce a…

★ 1.6kGrade B
fact-check✓SAFE

elvisun/newsjack

Extract factual claims from PR copy, verify each claim independently, attach concrete citations, and warn when certainty is low. Runs each claim through proven newsroom verification methods (lateral reading, source-tier…

★ 1.6kGrade B

Add this badge to your README

[![ModelsAtlas: SAFE B](https://modelsatlas.com/skills/elvisun/meanest-editor/badge.svg)](https://modelsatlas.com/skills/elvisun/meanest-editor/)

Questions about meanest-editor

What does the meanest-editor skill do?
The open-source skills that turn your agent into a full PR team.
Is meanest-editor safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can meanest-editor access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
How current is this page?
The grade is for one exact copy of the source (384ad36be393), read on 2026-10-28. The repository is watched, and a new audit runs when it changes — this is the first audit.

Sources

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.