---

name: content-maintenance description: "Helps maintain NLP-progress Markdown task pages by adding or updating datasets, benchmark results, SOTA rows, code links, table conventions, validation checks, and optional Jekyll previews." disable-model-invocation: true metadata: disco-role: operating license: MIT

content-maintenance

Use this sub-skill when a user asks to add or update NLP-progress task pages, dataset sections, benchmark result rows, SOTA ordering, paper/source citations, code implementation links, or a local site preview for Markdown content.

NLP-progress is a static Markdown and GitHub Pages corpus. Core content maintenance is CPU/any only and does not require an ML runtime, package install, accelerator, or network access.

Route elsewhere

• Automated JSON export, structured parsing, or structured.json generation belongs to structured-export.
• Domain benchmark discovery, benchmark lookup, or deciding what the current SOTA should be from external sources belongs to benchmark-catalog.

Operating procedure

1. Identify the target language/task Markdown page and whether the change is a result row, dataset section, or new task page.

2. Apply the bundled contribution policy in references/contribution-guidelines.md.

3. Preserve table conventions from references/markdown-table-style.md, including Model, metric columns, Paper or Paper / Source, and optional Code cells.

4. For new datasets, include description, evaluation setting/metric, annotated example, download link when available, and at least two results including the SOTA result.

5. Keep result tables sorted with the best result on top, while checking metric direction before moving rows.

6. Validate changed Markdown with the bundled checker before handoff. From the generated nlp-progress skill root, run:

   python3 sub-skills/content-maintenance/scripts/check_nlp_progress_markdown.py <changed-file-or-directory>
   python3 sub-skills/content-maintenance/scripts/check_nlp_progress_markdown.py --strict <changed-file-or-directory>
   

   From this sub-skill directory, the equivalent path is python3 scripts/check_nlp_progress_markdown.py ....

7. If the user explicitly asks for a rendered preview, use the optional Jekyll notes in references/site-preview.md. Do not make Ruby/Bundler setup part of normal content validation.

8. If validation or rendering fails, use references/troubleshooting.md to triage contribution, editor, build, and table issues.

Handoff checklist

• State which Markdown pages changed and what task/dataset/result each change affects.
• State the source basis for each new result: paper/preprint, benchmark split, metric name, and code link status.
• State how sorting was decided, including whether higher or lower metric values are better.
• Report checker command(s), warnings, and whether --strict passed.
• Mention any stale leaderboard caveat, missing download link, missing code link, or unavailable optional Jekyll preview.