Writing and editing

Use this skill when

• The user wants to draft, rewrite, or substantially restructure a document such as a proposal, spec, RFC, README, migration guide, onboarding doc, or runbook.
• The user wants a readability audit on already-written multi-paragraph prose.
• The user wants a short-string audit on metadata such as titles, descriptions, bios, taglines, schema descriptions, or README blurbs.
• Another skill produced human-facing prose or copy and the remaining work is writing quality rather than code or workflow design.

Do not use this skill when

• The main deliverable is an implementation plan, rollout plan, or reviewer-gated plan artifact — use plan-review.
• The main task is to sharpen a rough ask into a repository-grounded execution brief or contract-shaped prompt before work starts — use reverse-prompt.
• The main task is GitHub profile or repository surface setup rather than the writing itself — use github-presence.
• The output is primarily code, config, generated files, or a structured workflow handoff artifact rather than human-facing prose.

Inputs to gather

Required before editing

• Which mode applies: draft, readability-audit, or metadata-audit.
• The target document or strings.
• The intended audience and publishing surface.
• Any constraints on tone, length, or structure.

Helpful if present

• Existing notes, outline, or rough draft.
• Any paired fields that should be reviewed together, such as title + description.
• Whether the user wants review-only output or an actual rewrite.

Only investigate if encountered

• Whether a repo, product, or org-specific template should shape the document.
• Whether a broader repo-quality or profile-surface audit belongs to github-presence instead.

First move

1. Classify the task as draft, readability-audit, or metadata-audit.
2. Confirm the audience and target surface before changing wording.
3. Pick the matching reference file and stay inside that mode unless the user explicitly wants a cross-mode pass.

Workflow

1. Route to the right mode first.

   • draft: a structured document needs to be created or substantially reshaped.
   • readability-audit: the structure is mostly settled and the user wants prose-quality feedback.
   • metadata-audit: the task is about short audience-facing strings.

2. Draft mode.

   • Confirm document type and audience.
   • Propose an outline before drafting body sections.
   • Draft iteratively instead of front-loading the full document unless the user asks for a full first pass.
   • Run a fresh-reader pass once a complete draft exists.

3. Readability-audit mode.

   • Read the full prose before judging sentence-level issues.
   • Report concrete issues with quoted evidence and a Flesch Reading Ease score.
   • Keep domain terms that the intended audience would reasonably expect.

4. Metadata-audit mode.

   • Review strings in batches when possible.
   • Check front-loading, concreteness, filler, duplication, and truncation fit for the target surface.
   • Provide a concrete rewrite whenever a string fails an important check.

5. Handle cross-mode work deliberately.

   • Draft first, then run readability or metadata review only after the structure or field set is stable.
   • If the user really wants only one narrow pass, do not broaden into the other modes automatically.

Guardrails

• Must classify the task correctly before applying guidance.
• Must not treat short metadata strings as if they were full-document prose.
• Must not treat a settled readability audit as an excuse to redesign the whole document unless the user asks.
• Must not replace planning, prompt-sharpening, or GitHub-surface skills when those are the true deliverables.
• Should keep drafting iterative and outline-first.
• Should keep readability feedback evidence-based and specific.
• Should keep metadata feedback terse, concrete, and surface-aware.

Validation

• Confirm the chosen mode matches the actual deliverable.
• In draft mode, confirm the outline matches the document type and audience before filling body sections.
• In readability-audit mode, provide a Flesch score and quoted evidence.
• In metadata-audit mode, keep the output actionable in seconds and provide rewrites only where needed.

Reference files

• Read references/doc-types-and-boundaries.md when the task sounds like document work but may belong to another skill.
• Read references/review-loop.md when a complete draft needs a fresh-reader pass before handoff.
• Read references/readability-audit.md when running a prose audit on multi-paragraph text.
• Read references/metadata-audit.md when auditing titles, descriptions, bios, taglines, or similar short strings.