Simplified Technical English

Simplified Technical English

ASD-STE100 is a controlled language written so that a maintenance technician reading a second language can follow a procedure without misreading it. It has two halves: writing rules, and a dictionary that fixes one meaning per word.

This skill adopts part of the first half. It never ships the second.

The one-line version

An operator can get most of the value by naming the standard:

## Communication style
Use ASD-STE-100 when you speak to the operator.

That works because the name pulls a whole register the model already knows, at a cost of one line. Skill(imbue:latent-space-engineering) covers the technique in its named-register-invocation module, together with the known failure mode: practitioners report that the constraint decays partway through a long session. The checks below exist because guidance decays and counting does not.

What this skill is not

This is an STE-derived authoring aid. It is not an implementation of ASD-STE100, and no output of it may be called STE compliant. Read modules/licensing.md before writing anything that describes this work to a reader outside the project. The short version: the standard is free to download but not freely licensed, the controlled dictionary cannot be redistributed, and ASD states that it does not endorse or certify sellers of tools claimed to be fully compliant.

Vocabulary checking is therefore out of scope. Rules that depend on the approved-word list are not adopted and cannot be.

Decide the scope first

STE applies to some text in this repository and would damage the rest. Get this wrong and the 20-word cap starts deleting the reasoning that docstrings and ADRs exist to carry.

The last row is the one that bites. These skill files are descriptive repo prose, so the house rules govern them, not STE. Full reasoning and worked before-and-after pairs: modules/scope-boundaries.md.

The adopted rules

Four limits, corroborated across independent public restatements:

Plus: one instruction per sentence, active voice in procedures, simple tenses only, no contractions, no semicolons, and American spelling. Each restated in our own words in modules/adopted-rules.md, with the rules deliberately not adopted and why.

Running the checks

Three checks need counting or noun detection, so they are code:

cd plugins/scribe && PYTHONPATH=src uv run python -m scribe.ste FILE.md

Each finding prints as file:line: [rule/confidence] detail. The command exits 0 whenever it could read the files, because these checks are advisory and the noun-cluster rule must never gate a merge. Pass --no-noun-clusters to see only the two counting rules, which is the usual way to read a file that has never been checked before.

Four more are regex and live in the language pack, gated off so a routine slop sweep never runs them:

cd plugins/scribe && uv run python -c "
import sys; sys.path.insert(0, 'src')
from scribe.pattern_loader import get_ste_patterns, load_language_patterns
for e in get_ste_patterns(load_language_patterns('en'), include_optional=True):
    print(e['category'], e['confidence'], len(e['patterns']), 'patterns')
"

Mask code, tables, and frontmatter before you run the regex set. The scribe.ste checks already do this.

Read the findings honestly

Measured across 5984 markdown files in this repository:

A noun-cluster finding is advisory. Detection subtracts function words and verb forms and calls what is left a noun, which a real part-of-speech tagger would do better. Reread the phrase it points at. Do not rewrite on it, and never gate a merge on it.