techdocs-authoring
A Claude Code plugin that adds a technical-writer subagent and a
documentation-authoring reference skill — portable, vendor-neutral, and
conformant with the open Agent Skills specification.
It is built for writing and reviewing technical docs: picking a canonical
style guide, classifying each doc with the Diataxis four-mode model, applying
templates and reviewer checklists for READMEs, how-tos, tutorials, reference,
ADRs, runbooks, and release notes, single-sourcing content, running a disciplined
writing process, and making docs serve both humans and AI agents. The subagent
drafts and reviews in its own context, so doc-heavy work doesn't flood your main
session.
What's inside
| Component | Name | Purpose |
|---|
| Subagent | technical-writer | Drafting/review persona with a scoped tool allow-list; preloads the reference skill and works in its own context |
| Skill | authoring-reference | Portable writing craft — style-guide selection, Diataxis, templates/checklists, single-sourcing, the writing process, and dual human+AI-agent docs (progressive disclosure via references/) |
The skill conforms to the open Agent Skills
specification, so it is portable to other skills-compatible agents.
Install
/plugin marketplace add pkloehn1/techdocs-authoring
/plugin install techdocs-authoring
Or test locally without installing:
claude --plugin-dir /path/to/techdocs-authoring
Then delegate to it (e.g. "use the technical-writer agent to draft the
deployment runbook") or open the reference with
/techdocs-authoring:authoring-reference.
Prerequisites
- None hard. A docs-as-code workflow (Markdown in version control, reviewed
by pull request) is assumed.
- Optional:
WebFetch / WebSearch for pulling canonical style-guide pages
while writing.
Relationship to your repository's docs standards
This plugin carries portable craft; your repository keeps its
enforcement. The skill never restates repo-specific rules — line-length
limits, numbered-heading validation, diagram linting, pre-commit/CI wiring, or
repo paths — and defers those numbers and tools to your repo. Where a craft
principle here overlaps one your repo already adopts (Strunk-style economy, a
grade-9 readability target, single-source-of-truth), they are the same baseline,
not competing rules. Adopt the skill alongside your existing standards, not in
place of them.
Security & scope
- The content is vendor-neutral and ships no secrets.
- The subagent has a scoped tool allow-list so it can author and review docs
and run the project's Markdown linters, but cannot reach infrastructure or
mutation tooling. It authors prose only — it does not edit code, config, or CI.
The allow-list is
Read, Write, Edit, Grep, Glob, Bash(git:*), and
the doc linters Bash(pre-commit:*), Bash(markdownlint:*),
Bash(markdownlint-cli2:*), Bash(npx:*), Bash(vale:*), plus WebFetch and
WebSearch.
Versioning
Releases follow SemVer via the version field in
.claude-plugin/plugin.json. Bump it to ship an update; see CHANGELOG.md.
License
MIT.
