Journalism Agent Skills - Setup Guide¶
Source: jamditis/claude-skills-journalism via skills.sh - 18.9K combined installs across 67 indexed listings; first seen Oct 9, 2026 (extended-query probe)
GitHub: jamditis/claude-skills-journalism (416 stars, MIT license; active - pushed Oct 4, 2026; default branch master; 63 SKILL.md files across plugin packages and root-level skill packages)
Category: Journalism / Research / Media
Quality Tier: 🟡 Beta - 416-star MIT collection; docs site skills.amditis.tech; active Oct 2026; mixed sampled verdicts
James Amditis publishes one of the broadest journalism skill collections on skills.sh: 67 indexed listings covering verification, public records, data journalism, newsroom style, editorial workflow, crisis communications, academic writing, and media production. The same repository serves Claude Code and Codex, keeping Claude-only commands, agents, and hooks clearly labeled while shared skills follow the Agent Skills specification. Documentation lives at skills.amditis.tech, including setup guides and long-form workflow guides on autonomous dev work, multi-agent workflows, and persistent sessions.
The skills ship as plugin packages: journalism-core bundles 15 core skills (AP-style writing, source verification, FOIA and NJ OPRA requests, fact-checking, interview prep and transcription, story pitches, editorial workflow, crisis communications, newsletter publishing, photo metadata), research-toolkit adds six research and academic skills, dev-toolkit covers development work including ethical web scraping, security-toolkit handles defensive security, and smaller packages cover video, PDF design, project templates, and knowledge-base scaffolding. Root-level skill packages (okf-wiki, pdf-design, visual-explainer, web-design-picker) install on their own.
Installation¶
Use one install path per client; during the Codex pilot, stick to a single Codex installation path per skill or package.
Claude Code plugins (recommended in the README) also add slash commands. Prerequisite: Claude Code installed (claude --version to check).
/plugin marketplace add jamditis/claude-skills-journalism
/plugin install journalism-core@claude-skills-journalism
Codex can install the journalism-core package through the same marketplace metadata. Prerequisite: Codex CLI installed (codex --version to check).
codex plugin marketplace add jamditis/claude-skills-journalism
codex plugin add journalism-core@claude-skills-journalism
The skills CLI installs standards-based skills into the agent's skills directory. Codex example from the README:
npx skills@latest add https://github.com/jamditis/claude-skills-journalism/tree/master/journalism-core --skill '*' --agent codex --copy -g -y
To install a single skill instead:
npx skills@latest add jamditis/claude-skills-journalism --skill fact-check-workflow --agent codex --copy -g -y
Manual route: clone the repo and copy or symlink a skill directory into your agent's skills folder. Claude Code discovers skills at ~/.claude/skills/<skill-name>/SKILL.md, one level deep:
git clone https://github.com/jamditis/claude-skills-journalism.git
cp -r claude-skills-journalism/dev-toolkit/skills/web-scraping ~/.claude/skills/
What It Provides¶
Install counts from the Oct 10, 2026 skills.sh snapshot:
| Skill | Installs | Use For |
|---|---|---|
| web-scraping | 6,063 | Ethical scraping patterns: Playwright, robots.txt, anti-bot defense, and terms-of-service discipline (dev-toolkit) |
| academic-writing | 2,859 | Research design, literature reviews, IMRaD structure, peer review responses, and grant proposals (research-toolkit) |
| page-monitoring | 596 | Change detection, RSS generation, webhook alerts, and automatic archiving when monitored pages change (research-toolkit) |
| fact-check-workflow | 545 | Claim extraction, evidence gathering, rating scales, and correction protocols (journalism-core) |
The remaining 63 indexed listings range from 5 to 484 installs.
Why This Matters for Hermes Agents¶
Journalism is the discipline of checking things before publishing, which is exactly the discipline most agent workflows lack: sources get cited without verification, claims get repeated without extraction, and research trails go undocumented. This collection encodes the checking habit as skills - SIFT-based source verification, deepfake and C2PA checks, FOIA requests with current statutory citations, fact-check rating scales, and correction protocols - so an agent doing research work can follow a newsroom process instead of an improvisation. The scope reaches well beyond newsrooms: interview transcription pipelines, web archiving for evidence preservation, page-change monitoring, data analysis with chart and map generation, and academic writing all apply to any serious research task an agent runs. The repo also ships multi-agent workflow guides drawn from real newsroom practice (ICIJ, The Markup, Full Fact, Elicit) covering fan-out, pipeline, and adversarial-verify patterns. It is MIT licensed and maintained with a visible catalog control plane and validation scripts, and it serves Claude Code and Codex from one repository. For Hermes agents that research, monitor, and write, this is a strong verification layer to install alongside general-purpose content skills.
Usage¶
| You say | What happens |
|---|---|
| "Draft a FOIA request for these records" | foia-requests writes a federal FOIA or NJ OPRA request with current citations, then plans tracking and appeals |
| "Verify this video before we publish it" | source-verification runs the SIFT method with image/video verification, deepfake checks, and C2PA Content Credentials |
| "Fact-check the claims in this draft" | fact-check-workflow extracts claims, gathers evidence, applies a rating scale, and proposes corrections |
| "Analyze this dataset for a story" | data-journalism handles dataset analysis, charts and maps, statistical reasoning, and story structure |
| "Rewrite this headline in AP style" | newsroom-style enforces AP Style, attribution rules, headline formatting, and number conventions |
| "Watch this government page for changes" | page-monitoring detects changes, generates RSS or webhook alerts, and archives the page automatically |
| "Transcribe and quote-check my interview" | interview-transcription runs Whisper/WhisperX pipelines with quote management and speaker diarization |
Verification¶
# Claude Code: plugin skills land under the skills directory
ls ~/.claude/skills/ | grep -i -E "verification|foia|fact-check"
# Codex: user-level install target used by the skills CLI route
ls ~/.agents/skills | grep -i fact-check
# Review a skill file straight from GitHub before installing (default branch: master)
curl -s https://raw.githubusercontent.com/jamditis/claude-skills-journalism/master/dev-toolkit/skills/web-scraping/SKILL.md | head -20
The README also documents a clean-install canary that checks the exact global destination for the Codex route.
Security¶
skills.sh verdicts for sampled skills (verified Oct 10, 2026) - re-check the security pages on skills.sh before production use:
| Skill | Gen Agent Trust Hub | Socket | Snyk |
|---|---|---|---|
| web-scraping | Pass | Pass | Warn |
| data-journalism | Pass | Pass | Pass |
| fact-checking | - | - | - |
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.
Limitations¶
- Mixed security profile: web-scraping carries a Snyk Warn in the sampled verdicts, and sampling is uneven across the collection (fact-checking returned no verdicts).
- The Codex package route exposes nested skills only - Claude commands, agents, hooks, and root-level skills are not converted into Codex components.
- Codex installs are pilot-stage: the README warns against mixing installation paths for the same skill identity because Codex does not deduplicate same-name skills across install roots.
- Default branch is master, so raw GitHub URLs and manual clone flows must use master rather than main.
- Skills cite statutes, platform APIs, and vendor behavior that change; the README recommends re-checking older skills against their sources, with last-changed dates on the docs site.
- Snapshot data, verified Oct 10, 2026: 18,912 combined installs across 67 indexed listings; 416 GitHub stars; MIT; last pushed Oct 4, 2026. Counts drift over time.
Related¶
- Academic Research Skills - Paper Pipeline for Agents Setup - adjacent research pipeline for long-form academic work
- Bright Data Agent Skills - Web Scraping & Research Setup - production-grade scraping to pair with the ethical-scraping skill
- Research Paper Writing Pipeline - Academic ML/AI Paper Production Setup - writing pipeline for research outputs
- Skills Catalog | Skills Marketplace
Pro Tips¶
- Start with the journalism-core plugin: one install brings all 15 core skills, and the README treats it as the default entry point.
- Prefer the skills CLI route for single skills (for example
fact-check-workflow) when you do not want an entire plugin package. - Use one Codex installation path per skill: the repo warns that duplicate identities across install roots are not deduplicated.
- Do not clone the repository directly into
~/.claude/skillsas a nested folder; copy or symlink one level deep so Claude Code discovers each SKILL.md. - Check the last-changed dates before relying on statutory or platform-citing skills, since the repo tracks changes by git history.
← Skills Catalog | Skills Marketplace → Powered by CorpusIQ