ai-newsroom
Five blind editors and a chief editor who returns one fix list and a SHIP, FIX or REWRITE verdict.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/mavrin-xv--ai-newsroom)HTML
<a href="/listings/mavrin-xv--ai-newsroom"><img src="/api/listings/mavrin-xv--ai-newsroom/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Every script runs offline on Northwind Ledger, an invented bookkeeping company. `draft-v1.md` is flawed on purpose.
python3 skills/newsroom-review/scripts/structure_lint.py examples/northwind-ledger/draft-v1.md \ --profile examples/northwind-ledger/brand-profile.md --format guide
structure_lint: draft-v1.md (format guide) | Item | Result | Line | Detail | |---|---|---|---| | title | FAIL | 1 | 81 chars (max 70); banned opener | | description | FAIL | 1 | 169 chars (max 155) | | summary block | PASS | 10 | 4 bullets (want 3-5) | | headings | FAIL | 17 | 6 H2 (max 6, summary block not counted); banned: line 17 'Introduction', line 49 'Conclusion' | | template | FAIL | - | missing: When this does not apply | | hook | FAIL | 8 | opens with 'Month-end close is the' | | code and media | FAIL | 35 | code without language at [35] | | paragraphs | FAIL | 19 | line 19: 9 sentences (max 4) | | length | FAIL | - | 437 words (band 700-1600) | | close | FAIL | 49 | last H2 'Conclusion' | | cta count | PASS | - | 1 links to the CTA url (want 1-2) | | punctuation and padding | FAIL | 19 | '!' at lines [19]; padding: Additionally (line 19), It is worth noting (line 19) | | banned words | FAIL | - | found: seamless, best-in-class | 11 FAIL, 2 PASS
The same draft after three review rounds (`draft-v3.md`) gives `0 FAIL, 13 PASS`, and `claims_scan.py` goes from `9 claims, 7 not sourced` to `5 claims, 0 not sourced`. Every round is in the folder: `run-1/` to `run-3/` hold the briefs, the five verdicts, `merged.md` and the chief editor's `review.md`.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "ai-newsroom",
- "plugins": [
- {
- "name": "ai-newsroom",
- "source": "./",
- "description": "An editorial desk as agent skills: five blind editors (format, structure, voice, reader fit, claims) and a chief editor who returns one fix list and a SHIP, FIX or REWRITE verdict, plus an article producer and an X long-form writer, all driven by one brand profile.",
- "version": "0.1.0"
- }
- ],
- "owner": {
- "name": "Artem Mavrin",
- "url": "https://github.com/mavrin-xv"
- }
- }
README.md · README · 149 lines
- 
- # ai-newsroom
- **An editorial desk for your AI agent. Five editors who cannot see each other review a draft for format, structure, voice, reader fit and claims. A chief editor merges their notes into one fix list and a SHIP, FIX or REWRITE verdict. Two writing skills sit on top: one for blog articles, one for X long-form. Every rule comes from one brand profile you fill once.**
- An AI draft usually reads fine, and that is the problem. It takes a careful reader to spot the figure with no source, the title that names a topic instead of an outcome, the sentence the author would never say, or the slug nobody checked.
- [](LICENSE)
- 
- 
- 
- ```bash
- git clone https://github.com/mavrin-xv/ai-newsroom && cd ai-newsroom
- ./install.sh # or: ./install.sh --project
- ```
- <!-- install-notes:start -->
- Other ways in: as a Claude Code plugin, run `/plugin marketplace add mavrin-xv/ai-newsroom` and then `/plugin install ai-newsroom@ai-newsroom`. For Codex or another agent that reads `SKILL.md`, copy the folders under `skills/` into its skills directory. On Windows, run `install.sh` from Git Bash or WSL and type `py -3` where these docs say `python3`.
- <!-- install-notes:end -->
- Then tell your agent: `set up my brand for the newsroom`, then `write a guide on <topic>` or `run the editors on draft.md`.
- ---
- ## What it solves
- | Pain | What the pack does |
- |---|---|
- | One reviewer, one blind spot | Five editors in separate contexts, each with its own lane and only the profile sections that lane needs; none sees another's verdict |
- | Five opinions, no decision | `review_run.py merge` dedupes findings across editors, checks every quoted line is really in the draft, proposes a verdict; the chief editor resolves conflicts by a fixed order of precedence |
- | Figures nobody can trace | `claims_scan.py` lists every number, superlative and absolute and matches it against the Proof table and `sources.md`; the claims editor audits the rest |
- | "It reads fine" but breaks the house rules | `structure_lint.py` measures title, description, summary block, H2 count, required sections, hook, paragraph length, word band, CTA count and banned words from the profile |
- | The author's voice drifts into generic AI | Voice profiles built from 5 or more real samples; the voice editor grades against them, not against taste |
- | Review rounds that never end | Each round is merged against the last (`--prev`): resolved, still open, new; a minor ledger that never re-raises; a round cap |
- | Guessed slugs and cut-off descriptions | `link_check.py` checks links against an allowed list; `seo_fields.py` measures title, description, slug and variants |
- | The blog post pasted into X | `x-article` picks one of five formats by intent and `x_article_check.py` checks title length, 5:2 cover, paragraph length, links per paragraph and CTA count |
- ## What's inside
- | Skill | What it does | What to say |
- |---|---|---|
- | `newsroom-setup` | Interview plus your existing material into `brand-profile.md`; voice profiles from samples; inferred lines marked for you to confirm | "set up my brand", "add an author voice" |
- | `newsroom-review` | Lints, blind briefs, five parallel editors, merge, chief editor, loop until SHIP or round 5 | "review this draft", "is this ready", "run the editors" |
- | `article-producer` | Brief, sources, draft, lint, the review loop, then the publish package: fields, links, alt text, export | "write a guide on", "turn these notes into an article" |
- | `x-article` | Five formats, seven levers, a source ladder, a six-part skeleton, the measurable checks | "write an X article", "turn this post into X long-form" |
- ## Quick start
- Every script runs offline on Northwind Ledger, an invented bookkeeping company. `draft-v1.md` is flawed on purpose.
- ```bash
- python3 skills/newsroom-review/scripts/structure_lint.py examples/northwind-ledger/draft-v1.md \
- --profile examples/northwind-ledger/brand-profile.md --format guide
- ```
- ```
- structure_lint: draft-v1.md (format guide)
- | Item | Result | Line | Detail |
- |---|---|---|---|
- | title | FAIL | 1 | 81 chars (max 70); banned opener |
- | description | FAIL | 1 | 169 chars (max 155) |
- | summary block | PASS | 10 | 4 bullets (want 3-5) |
- | headings | FAIL | 17 | 6 H2 (max 6, summary block not counted); banned: line 17 'Introduction', line 49 'Conclusion' |
- | template | FAIL | - | missing: When this does not apply |
- | hook | FAIL | 8 | opens with 'Month-end close is the' |
- | code and media | FAIL | 35 | code without language at [35] |
- | paragraphs | FAIL | 19 | line 19: 9 sentences (max 4) |
- | length | FAIL | - | 437 words (band 700-1600) |
- | close | FAIL | 49 | last H2 'Conclusion' |
- | cta count | PASS | - | 1 links to the CTA url (want 1-2) |
- | punctuation and padding | FAIL | 19 | '!' at lines [19]; padding: Additionally (line 19), It is worth noting (line 19) |
- | banned words | FAIL | - | found: seamless, best-in-class |
- 11 FAIL, 2 PASS
- ```
- The same draft after three review rounds (`draft-v3.md`) gives `0 FAIL, 13 PASS`, and `claims_scan.py` goes from `9 claims, 7 not sourced` to `5 claims, 0 not sourced`. Every round is in the folder: `run-1/` to `run-3/` hold the briefs, the five verdicts, `merged.md` and the chief editor's `review.md`.
- ## How it works
- ```
- brand-profile.md ──┐
- draft.md ──────────┤
- ▼
- structure_lint.py + claims_scan.py (measure, do not judge)
- │
- review_run.py packet (one brief per editor, only its sections)
- │
- ┌────────┬───────┼────────┬─────────┐
- format structure voice icp claims (parallel, blind to each other)
- └────────┴───────┼────────┴─────────┘
- review_run.py merge --prev (quotes checked, deduped, verdict proposed)
- │
- chief editor (conflicts resolved, one fix list)
- │
- SHIP ──► publish package FIX ──► apply all, next round REWRITE ──► new draft
- ```
- The scripts do the counting, so the editors spend their attention on judgement. The chief editor may move the proposed verdict one step and must say why.
- ## Configure
- Everything brand-specific lives in `brand-profile.md` (template in `templates/`). Lines of the form `- key: value` are read by the scripts.
- | Setting | Where | Default |
- |---|---|---|
- | Word band, max H2, summary block, required sections | `## Formats`, one `###` block per format | 800-2000 words, 6 H2 (without a profile) |
- | Title and description length, sentences per paragraph | `## Channels` | 70, 155, 4 |
- | CTA url and how many times per article | `## CTA` (`url`, `max`) | none, 2 |
- | Banned words | `## Banned` | none |
- | Allowed links | `## Links` | none |
- | Recommended findings that still allow SHIP | `review_run.py merge --recommended-limit` | under 5 |
- | Blocking findings that force REWRITE | `review_run.py merge --rewrite-at` | 3 |
- | Round cap | `review_run.py merge --max-rounds` | 5 |
- ## Honest limits
- - The example ran three real rounds and ended at FIX, not SHIP. Blocking items fell from 14 to 0, but each round the editors raised about a dozen new recommended items, above the limit of 5. Either relax `--recommended-limit` for your team or expect the round cap to be what ends the loop.
- - Editors follow the profile's prose, not the scripts. When a rule lived only in a script (the summary block does not count as an H2), the editors counted it anyway until the sentence went into the profile.
- - `claims_scan.py` is a pattern matcher. It flags some instructions as superlatives and misses claims phrased without a number. The claims editor is the real audit.
- - Nothing here checks facts against the world. A figure is "sourced" when it has a row; whether the row is true is yours to verify.
- - Tested with Claude Code subagents on Windows. Codex and other agents, and macOS and Linux, are not tested.
- - X platform facts were checked in 2026-11 and will drift.
- ## Credits
- - Based on / inspired by https://github.com/obra/superpowers (MIT): fresh-context subagents per task, review before you trust, verification before claiming done
- - Based on / inspired by https://github.com/anthropics/skills (Apache-2.0), doc-coauthoring: reader testing in a fresh context, one section at a time
- - Based on / inspired by https://github.com/coreyhaines31/marketingskills (MIT): a product context file every skill reads, copy-editing sweeps by lane
- - Based on / inspired by https://github.com/wshobson/agents (MIT) and https://github.com/VoltAgent/awesome-claude-code-subagents (MIT): single-lane reviewer agents with fixed output contracts
- - Based on / inspired by Cornell University Marketing and Communications, content accuracy checklist: https://help.marcom.business.cornell.edu/review/content-accuracy (checklist idea; no text copied)
- - X Help Center, Articles: https://help.x.com/en/using-x/articles (platform facts restated)
- - Format and lever ideas drawn from public X long-form guides by Typefully, OpenTweet and PostEverywhere (ideas only; no text copied)
- ## Contributing
- See [CONTRIBUTING.md](CONTRIBUTING.md). Notes for coding agents are in [AGENTS.md](AGENTS.md).
- <!-- author:start -->
- ## Author
- Built and maintained by [Artem Mavrin](https://github.com/mavrin-xv). This repo is one of a set of agent skill packs for marketing and operations work. Each one installs on its own; the full list is in [agent-skill-packs](https://github.com/mavrin-xv/agent-skill-packs). If a pack saved you an afternoon, a star helps other people find it.
- <!-- author:end -->
- ## License
- MIT. See [LICENSE](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.