Add journal-intelligence skill: live guidelines, AI policy, LaTeX template fetch

[eniktab]

Summary

Adds a journal-intelligence skill that fetches live author guidelines, AI authorship policy, and LaTeX template for any journal before writing — filling a gap in the current scientific-writing skill, which uses hardcoded format requirements that go stale.

What it adds

• Live WebSearch/WebFetch of author guidelines for any journal (Nature family, Elsevier, Wiley, Cell Press, PLOS, Springer, ACS, RSC, IEEE, ACM, and more)
• Structured extraction into journal_profile.yaml (20+ fields: word limits, abstract format, figure limits, methods placement, citation style, required special sections)
• Separate fetch of AI authorship policy with exact required disclosure language and placement
• LaTeX template discovery and download (Overleaf gallery, publisher template hubs)
• Editorial priority and desk-rejection criteria lookup
• Fallback with [VERIFY] flags when network is unavailable

Why hardcoded limits are a problem

What changed	When
Nature Biotechnology article length	2023
AI authorship disclosure (all major publishers)	2023–2025 (ongoing)
LaTeX template versions	Continuously
PLOS data availability mandate	2024

Submitting with stale requirements causes desk rejection. The fix is always fetch live.

Integration with existing scientific-writing skill

journal-intelligence runs as Phase 0 before scientific-writing. The journal_profile.yaml it produces is consumed directly to set word limits, citation format, required sections, and template class — overriding defaults.

Full standalone repo

https://github.com/eniktab/claude-scientific-paper-writer

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e919f83de1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Add this skill to the marketplace skills list

This commit creates journal-intelligence, but the plugin manifest still uses an explicit allowlist of skill directories in .claude-plugin/marketplace.json and does not include ./scientific-skills/journal-intelligence. In that packaging flow, the new skill is not discoverable/installable even though this file exists, so users cannot actually invoke the feature being introduced.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Remove or implement the claimed auto-integration

This line states that scientific-writing directly consumes journal_profile.yaml, but there is no corresponding reference to journal-intelligence or journal_profile.yaml in scientific-skills/scientific-writing/SKILL.md (repo-wide search only finds these terms in this new file). As written, users can reasonably assume automatic Phase-0 execution and skip an explicit run, which means the live guideline step may be missed.

Useful? React with 👍 / 👎.

[eniktab]

Thanks for the review — both points addressed in the latest commit (e77b2fd):

P1 — marketplace.json: Added ./scientific-skills/journal-intelligence to the skills allowlist in .claude-plugin/marketplace.json, placed alongside scientific-writing and citation-management.

P2 — Auto-integration claim: Removed the misleading statement. The section now explicitly documents a two-step manual workflow — run journal-intelligence first to produce journal_profile.yaml, then reference that file explicitly when invoking scientific-writing. The two skills remain independent with no implicit chaining.

[Gonzih]

Overall this is a well-structured and genuinely useful skill — the motivation (hardcoded journal requirements go stale) is sound, and the journal_profile.yaml schema captures the right fields. A few things to address before approval:

Needs clarification / fixes

1. allowed-tools in frontmatter is non-standard for this repo. The existing skills do not use an allowed-tools key in YAML frontmatter — this appears to be a field specific to your standalone repo. Check whether the K-Dense marketplace format supports or ignores this field. If it's ignored, it's harmless clutter; if it causes a parse error, it needs to be removed.

2. WebSearch/WebFetch dependency is a hard runtime requirement. This skill only works if the agent runtime has WebSearch and WebFetch tools available. The fallback section acknowledges network failures but not the case where these tools simply aren't configured. A brief note under "When to Use" that this skill requires WebSearch and WebFetch to be enabled in the agent's tool set would set accurate expectations for users.

3. URL patterns are partially outdated or publisher-specific. The listed URL patterns for Elsevier (/journals/[journal]/[issn]/guide-for-authors) and PLOS (/journals/[journal]/s/submission-guidelines) are reasonable starting points, but Elsevier in particular has a highly variable URL structure. Framing these as "example patterns to try" rather than authoritative patterns would be more accurate.

4. The self-promotion link at the bottom (Part of the claude-scientific-paper-writer suite) — maintainers may want to remove or reframe this, as other contributed skills don't link to external contributor projects from within the skill body.

What's good

• The journal_profile.yaml schema is comprehensive and well thought out — the 20+ fields cover the full spectrum of what a writer actually needs.
• Separate ai_policy.md output for AI disclosure language is a smart design choice given how rapidly those policies are changing.
• The two-step workflow description (journal-intelligence → scientific-writing) with explicit handoff via the YAML file is clear and practical.
• The fallback behavior with [VERIFY] flags is a responsible approach.
• Correct registration in marketplace.json, placed logically before scientific-writing.

Addressing points 1–2 would make this ready to approve.

[Gonzih]

Solid, well-motivated skill. The core insight — that journal requirements change frequently enough that hardcoding them is brittle — is well-supported by the rationale table (Nature Biotech 2023, AI policy changes 2023–2025, PLOS 2024). Making guideline fetch a discrete Phase 0 before scientific-writing is the right architectural decision.

Specific things that work well:

• The known URL pattern table for major publishers (Nature, Elsevier, Cell Press, PLOS, etc.) gives Claude a fast-path before falling back to WebSearch.
• Storing results as journal_profile.yaml with 20+ structured fields (word limits, methods placement, citation style, AI disclosure) makes the output machine-readable for downstream steps.
• The AI policy extraction step (ai_policy.md) is timely given how rapidly publisher AI policies are evolving.
• Fallback behavior with [VERIFY] flags when network is unavailable is the right safe default.
• Tool dependencies (WebSearch + WebFetch) are clearly called out upfront.

Minor suggestions (non-blocking):

• Consider adding a brief ## Example Output snippet showing what a populated journal_profile.yaml looks like — this would help users understand the schema before running the skill.
• The marketplace.json placement between scientific-critical-thinking and scientific-writing is logical; just verify the maintainer is OK with the insertion point.

LGTM overall.