Executable bundle proposal: JP Essay writing workflow for Claude Code

[sadarume]

I am submitting this on behalf of the JP Essay project team, with AI assistance, using the executable-bundle proposal route described in CONTRIBUTING.md.

Source: https://github.com/lucasbabur/jp-essay

Revision reviewed: aa5f8991d0593c00966fe257816b77dae2fb9a90 (package version 0.2.0, MIT).

Proposed capability

JP Essay is a Claude Code writing workflow: question, research, outline, draft, a separate reader's reconstruction of the argument, structural revision, sentence revision, fact-checking, and a final reader pass. It produces one final essay per invocation and retains intermediate Markdown files.

Example input: /jp-essay Why do we procrastinate on things we care about? 1200 en.

Expected result: one essay plus inspectable source notes, outlines, drafts, reader reports, and a fact-check report under ~/Documents/essays/<slug>/. The checker validates mechanical requirements; it does not establish writing quality or factual correctness on its own.

Fit and overlap

• avoid-ai-writing focuses on auditing and rewriting existing prose; JP adds a research-to-final-essay workflow and intermediate artifact checks.
• scientific-writing focuses on scientific manuscripts and reporting conventions. The proposed initial entry is JP's general-reader essay skill, not a replacement for manuscript preparation.
• crossframe-essay is explicitly scoped to CrossFrame and Chinese-canonical writing. JP's documented output languages are English and Portuguese, with no CrossFrame dependency.

Complete proposed initial bundle

One canonical jp-essay tree containing the upstream SKILL.md, check.py, style.md, tells.md, and the full MIT LICENSE notice. The upstream test_check.py provides checker regression tests and would also need the appropriate maintainer-approved placement. Preserve upstream authorship, pin the source revision, and add the required source/license metadata.

The scientific entry point is not part of this initial proposal, avoiding a second skill sharing the same core. The icon, website, npm installer, and shell installer are not needed in the imported bundle and would not be copied.

check.py is executable Python with a uv script header requiring Python 3.11+, pysbd==0.3.4, and httpx==0.28.1. This is why I am proposing the complete bundle here before opening a fork PR. I will not omit the checker, disguise it as data, or submit an incomplete Markdown-only copy.

Integration and operational boundaries

• Claude Code is the supported host; no cross-agent compatibility is claimed.
• Upstream instructions currently hardcode ~/.claude/skills/jp-essay/check.py. An import would need a reviewed path adaptation compatible with your install layout, while preserving the gate behavior. No adaptation has been implemented yet.
• Claude Code/model access and uv/Python are required. Model and web-research use may incur costs. uv may download the declared Python dependencies.
• The writing agent creates and revises local essay artifacts. General essays can use user-supplied records; text given to Claude/subagents is processed through the user's configured model service.
• Research uses available web tools. The checker's source gate also fetches the source URLs with HTTP GET, following redirects; it is not an offline-only check. Network permission and private-source handling would need explicit treatment in the import.
• The workflow generates local files; it does not require publishing the essay or contacting anyone.
• JP is independent and is not affiliated with or endorsed by Jordan Peterson, whose essay-writing guide inspired the process.

Evidence and request

I inspected the upstream skills, checker structure/dependencies, installation code, documentation, and the nearby catalog entries above. This is a packaging proposal, not a completed import or a claim of passing your preflight. No new end-to-end model execution or independent security audit was performed for this submission.

Would a maintainer sponsor the protected review path for this single executable skill bundle? If so, please indicate the preferred checker/test placement and path-resolution convention before an implementation is prepared.

[sickn33]

Thanks for the detailed proposal. We reviewed the proposed revision aa5f8991d0593c00966fe257816b77dae2fb9a90 and think the research-to-final-essay workflow adds a useful capability beyond our existing editing skills. All 36 checker tests passed in our local evaluation.

Please open a draft integration PR for the single general-reader jp-essay bundle. Include SKILL.md, check.py, style.md, tells.md, the retained MIT LICENSE, and the checker regression tests, with the exact source revision and AAS provenance metadata. Please keep the scientific entry point, installers and website outside this initial PR. Keep the tests alongside the checker in the skill bundle so their relative file resolution remains valid.

Please address these adaptations in the draft:

• Resolve the checker relative to the installed skill instead of hardcoding ~/.claude/skills/jp-essay/check.py. Make the output destination explicit and protect existing essay files from accidental overwrites.
• Bound source URL fetching, including response size and redirect handling. Reject loopback/private-network destinations by default, with any explicitly authorized local-source use handled separately; provide an offline/manual verification path where network access is unavailable.
• Keep checker results clearly mechanical: a passing facts gate does not establish that every factual claim was checked or that the sources support it. Preserve the separate contextual verification step.
• Bound correction/retry loops and state the supported host and languages accurately. The proposed checker supports English and Portuguese; no general cross-host or Italian support should be implied.

Use the standard PR template and include the adaptation diff, test results, practical host-test evidence and any untested boundaries. Keep the change source-only, without generated catalogs or plugin mirrors.

The Python helper requires the protected maintainer review path described in CONTRIBUTING.md. Opening the draft lets us review the concrete candidate; the ordinary fork check cannot authorize its executable content, and the draft will need maintainer handling before it can merge. Please retain the complete helper rather than omitting it or renaming it as data.