When a release lands and the README needs updating, choose a writing mod that knows the job before asking a general coding helper to improvise. A focused helper can turn source changes into a changelog, check an API page, or edit prose. The median preload is 62 tokens, roughly a short paragraph on your screen.
Match the writing job to the helper
A skill is a folder of instructions the model opens when it decides it needs them. The context is everything the model can see while it answers you. Choose by subject, not by label.
This catalogue covers 17,683 writing mods. Some focus on API references, project documentation, or reviewing words shown to users. Choose the description that matches the work in front of you.
A token is the unit your AI plan bills by, about three quarters of a word. A narrow writing helper can keep its cost small while giving the model more relevant guidance.
What documentation mods cost before they write
The catalogue marks 17,176 mods as loading every session. Another 507 wait until you ask. The median preload is 62 tokens, and nine in ten cost 150 tokens or less. At 20 sessions a day, the 62-token median becomes 1,240 tokens across a day.
The heaviest recurring preload is tdoc, at 1,529 tokens every session across 1,306 lines. Its size matters most when its documentation task is rare.
What the writing scan catches
The scan graded 17,229 mods A, 246 B, 207 C or D, and 1 E or F. The most common finding types were 521 network calls, 75 recursive force deletes, and 73 reads of agent configuration directories.
docs-writer and record-gotcha received grade C for hidden instructions. code-wiki received grade C for recursive force deletion.
A grade A means the scan found nothing it treats as risky. Each finding points to a line you can inspect. You decide whether the mod belongs in your project.
Six starting points for docs and changelogs
- draft-release-notes: writes the release notes for a Prisma release by reading the merged pull requests. 174 tokens every session, 305 lines, original, 1 copy.
- write-api-reference: Formats concise reference pages for Next.js functions, components, file conventions, directives, and configuration. 130 tokens every session, 149 lines.
- adev-writing-guide: Applies technical-writing and Angular Markdown guidance to developer documentation. 60 tokens every session, 179 lines, 1 copy.
- code-documentation: Guides README files, API references, code comments, architecture notes, changelogs, and developer guides. 88 tokens every session, 416 lines, 4 copies.
- string-reviewer: Reviews interface labels, status messages, and error text for clarity and brevity. 38 tokens every session, 99 lines, 1 copy.
- docs-editor: Checks OpenAI Cookbook notebooks, Markdown, and MDX for accuracy, clarity, and publication readiness. 58 tokens every session, 85 lines.
When a rule or hook is the better tool
A glob is a filename pattern, like *.tsx. A rule with a glob suits guidance for matching files only. A hook is a shell command your agent runs for you when something happens. Use a hook when an action must happen on an event. This kind of writing mod cannot fire on one.
Wait for writing guidance when the task is occasional, which is the shape 507 of these mods already have. Use an event-driven tool when the timing has to be guaranteed.
Questions people ask
Should documentation helpers load every session? Only if you write docs in most sessions. The median body of an on-demand mod is 1,275 tokens when invoked, so occasional changelog work may favour waiting.
What should make me stop before installing one? A request for root access is a clear pause point. The scan recorded 59 such findings, so read the evidence line before deciding.
Does original status matter? The original is the first copy of a file we saw; everything matching it afterwards is a copy. There are 16,801 originals and no near duplicates. Provenance helps, but job fit and the scan grade should decide.
Do it now
Run npx agentmods cost to see what the mods in your current project cost per session, then browse the skills catalogue.