← ALL WORK

Article Agent.

Open source · v1.0.0

A writing toolkit for Claude: research, draft, revise, and plan. Local Python tools check prose while preserving code and structured content.

03 / ARTICLE AGENTRECORDED OUTPUT · EXCERPT
draft.md
BEFORE

Utilize the cache.

AFTER

Use the cache.

CODE IN BOTH VERSIONS
const mode = "utilize";

Change the prose. Keep the syntax.

One prose edit; code preservedLocal cleaner run · 18 Sep 2026
01 / THE PURPOSE

A technical draft is more than its sentences.

Code samples, links, and structured content sit alongside prose. Article Agent combines writing workflows for Claude with local Python tools for cleaning text and finding passages that need a closer edit.

WHO IT’S FOR
Writers working with Markdown, code, and Claude.
THE CONSTRAINT
A replacement that helps a sentence can break a code sample. Prose edits need a boundary around technical content.
02 / THE DECISIONS

Edit the sentence. Keep the syntax.

The cleaner separates protected spans from prose before applying its editing rules. Mechanical cleanup and suggestions for deeper revision remain distinct parts of the workflow.

  1. Separate the editable material

    The cleaner puts code, link targets, front matter, and math into a temporary vault before applying prose rules. It restores those spans after the edits.

  2. Leave the judgment with the writer

    Local rules handle mechanical cleanup. Structural patterns become suggestions for revision, while the writing workflow ends with a draft for the author to review.

03 / THE IMPLEMENTATION

Hide. Edit. Restore.

The sequence is the design decision. Prose passes work on masked text, then the saved technical spans return to the document.

skills/art-human/clean.pyPython
vault = Vault()
masked = vault.hide(text)

masked, e = stage_glyphs(masked, pat); edits += e
masked, e = stage_diction(masked, pat); edits += e
masked, e = stage_tidy(masked);         edits += e

restored = vault.show(masked).strip() + "\n"
return Outcome(restored, edits, find_shapes(restored, pat))
From the implementationExplore this source
Before editing
The vault masks technical spans so prose substitutions cannot rewrite their contents.
After editing
The original spans are restored. Structural findings become advice for human revision.
04 / WHERE IT STANDS
Open source · v1.0.0

Open source, from workflow to cleaner.

The v1.0.0 toolkit includes Claude writing skills, local Python utilities, regression-test source, and installation documentation. Start with a copy of a draft and review the changes before accepting them.

KEEP IN MIND

The writing workflows need Claude; local cleanup uses Python. A whole-document pass removes invisible characters, including inside code.

Take a closer look.

  1. Review the toolkit’s installation and workflow guide.
  2. Use the local cleaner on a copy of a Markdown draft, or explore the Claude writing skills.
  3. Read the changed lines and suggestions before accepting edits or publishing.
Explore the writing toolkitInstallation and workflow guide on GitHub
Sources, credits & review notes

Contribution & credits. Project author, credited in the repository and plugin manifest.

The recorded example uses revision 14e5cd9. Its source and pattern hashes, input, output, and edit list are included in the downloadable record. The local run was reproduced on 19 September 2026 and matched the stored result; Python does not execute in this page.

Invisible-character removal happens before masking and can change technical spans. The later prose-editing passes protect the masked material, rather than guaranteeing byte-for-byte preservation of the original file.

The full recorded sample preserved its inline code, fenced code, front matter, and link target. This is one example, not a general guarantee. The scoring tools use local heuristics, not a validated writing-quality metric or authorship detector.

PUBLIC SOURCE REVIEWED / 18 SEPTEMBER 2026
NEXT FROM THE WORKBENCH / 01KiroPad