Technical writing

Technical Writing

Patterns specific to documentation and developer-facing prose. Rules drawn from the legacy writing-guide, the Google Developer Documentation Style Guide, and Strunk & White.

Voice: Active vs Passive

Prefer active voice. It is shorter, clearer, and makes the actor explicit.

Passive (weak)	Active (strong)
"The cache was cleared by the scheduler."	"The scheduler clears the cache."
"Errors are handled by the retry middleware."	"The retry middleware handles errors."
"The API can be called with an optional token."	"Callers may pass an optional token."

Passive voice is acceptable when:

• The actor is unknown or unimportant: "The feature was deprecated in 2023."
• The object is the real topic and moving it to the subject position helps flow: "Passwords are never stored in plaintext."

In API docs, always state who does what: callers, the service, the library, the user.

API Documentation Patterns

A standard API reference entry has five parts:

1. One-sentence summary — what the function does, imperative present tense.
2. Parameters table — name, type, required?, description.
3. Return value — type and meaning.
4. Errors — what the function throws or returns on failure, and why.
5. Example — at least one runnable snippet per public function.

Keep each part terse. A reader arriving from a search result should see the summary without scrolling.

README Structure

A README for a developer tool should answer, in order:

Section	Question answered
Title + tagline	What is this?
Problem / motivation	Why does it exist?
Install	How do I install it?
Quick start	What does the shortest useful command look like?
Usage / commands	How do I actually use it?
Configuration	What knobs are there?
Development	How do I contribute?
License	What are the terms?

Not every section is required. But the order matters: readers bounce if they have to scroll past philosophy before seeing npm install.

Error Messages

An error message is prose, and its tone should be:

Dimension	Requirement
Voice	Active: "Config missing: run tool init to create one."
Blame	Blame the situation, not the user. Not "You forgot to …".
Specificity	Name the missing thing, the expected thing, or the fix
Length	One line when possible; multi-line only if a fix is needed
Capitalization	Start with a capital; terminal period optional
Punctuation	Avoid exclamation marks

Bad	Good
"Error!"	"Failed to parse config: expected {, got [."
"You're wrong."	"Port 6170 is already in use; pass --port to choose another."
"Something happened."	"Network request timed out after 30s."

Terminology Consistency

Pick one term per concept and use it everywhere in one document.

Do not mix	Within one doc, pick
function, method, routine, handler	one term, e.g. "function"
config, configuration, settings	one term
log in, login, sign in, sign-in	one pair: verb ("log in") + noun ("login")

Capitalize tool and framework names exactly as their owners do: React, TypeScript, npm, GitHub, PostgreSQL, iOS, macOS. This is a common L2 slip.

Headings

Headings state the topic, not the conclusion. In reference docs they are usually noun phrases.

Bad (full sentence)	Good (topic phrase)
"How you can install the tool"	"Installation"
"Here is what to do when it fails"	"Error handling"
"These are the configuration options"	"Configuration"

No terminal punctuation in headings.

Lists

• Parallel structure. Every bullet starts the same way (all imperative, or all noun phrases, or all full sentences).
• Use ordered lists when order matters (steps to follow) and unordered otherwise.
• Long bullet lists (more than seven items) usually hide a missing abstraction — consider a table.

Tables

Tables beat prose for reference data. Prefer tables when:

• Two or more dimensions vary (e.g. level × behaviour).
• The reader will scan rather than read.
• Alignment makes comparison easy.

Table columns need parallel headings. First column usually holds the key (metric, flag, level). Align numbers to the right.

Cross-References

Link the first mention of a concept, command, or component. Subsequent mentions in the same page usually do not need a link. Never write "click here" or "see below" as the link text — the linked phrase should read as the target name.

Concise Phrasing in Docs

The wordy → concise table from tone-calibration applies in full here. The corrections most common in docs:

Wordy	Concise
"in order to"	"to"
"at this point in time"	"now"
"has the ability to"	"can"
"is going to"	"will"

Scope

This skill covers technical writing patterns only. Related concerns live in sibling skills:

• grammar-fundamentals — general grammar independent of documentation
• punctuation-rules — punctuation conventions
• tone-calibration — per-context tone including commit, PR, email
• common-non-native-mistakes — recurring L2 patterns
• writing-guide — meta-skill routing to the four above