OneRead is a compiler for technical English, not a prompt. It type-checks against ASD-STE100 Issue 9 mechanical rules, then emits the same IR to docs, OpenAPI, SARIF, skills, MCP, Slack, and aviation procedures.
Input
Mechanical findings
Regex, not a grammar parser. Not a compliance verdict. Official standard: asd-ste100.org.
Compiled, mechanical subset
The compiled text appears here with every rewritten span marked. A span that stays is a span the compiler refuses to invent.
Paraphrased from ASD-STE100 Issue 9 with software examples. Issue 9 added no new rule numbers. It revised the wording of 31 rules. The official wording is in the free standard.
People have packaged STE as agent skills. A skill is still a prompt. The next call drifts. CI cannot fail a prompt. OneRead is the gate: parse, type-check, lower to ORIR, emit.
| Capability | Agent skill (prompt) | OneRead (compiler) | Authoring checkers |
|---|---|---|---|
| Agent skill / prompt | Yes | Yes, as an adapter | No |
| Deterministic linter | No | Yes, line-level | Yes, in the editor |
| Intermediate representation | No | ORIR 1.0 | No |
| Emit to OpenAPI / SARIF / Slack / MCP / AMM | No | Ten surfaces | Usually one editor |
| GitHub Action + SARIF | No | Yes | Rare |
| MCP server | No | Stdlib stdio | No |
| Interactive rule explorer | No | This page | Vendor UI |
| Official dictionary | Not reproduced | Not reproduced | Often licensed |
| Certifies STE | No | No | Still a human |
| CI can fail it | No | lint exit 1 | Editor only |
A skill can look clean on a good call. The next call puts check, verify, and ensure back in. CI cannot fail a suggestion. OneRead fails the pull request.
ORIR is JSON. Emitters read ORIR. They do not re-parse prose. That is the LLVM idea, applied to technical English. Every preview below compiles the text that is in the Playground input right now, so a change there changes all ten surfaces at once.
Each preview follows the shape of the Python emitter in src/oneread/emit.py. Run the CLI for the artifact itself.
ASD-STE100 Simplified Technical English is the controlled language aerospace and defense manufacturers use for maintenance documentation. Issue 9 is dated 15 January 2025. It is a free download from the STEMG site. It is an EU trademark of ASD. This playground is unofficial.
Models rotate synonyms, hedge with should, may, and could, bury the condition after the command, and write 40-word sentences. A non-native on-call engineer at 02:00 reads that once. STE was designed for that reader. Agent instructions are the same job: a reader that cannot ask a question.
STE was built for aircraft maintenance. OneRead emits an AMM-shaped procedure from the same IR you use for a README. The operational picture around that aircraft, delay, cascade, and safety-report retrieval, is a different product: AeroFarr, Zorost's aviation intelligence platform. Calibrated disruption forecasts, causal explanation, network cascade, and a cited safety corpus. Live at aerofarr.com.
OneRead does not certify compliance, and it does not reproduce the ASD dictionary, which is licensed. It rewrites only what a rule can prove: contractions, filler, Latin abbreviations, punctuation, and wordy phrases. A sentence split, a voice change, and a banned modal are reported with a rule number and left alone. A compiler that guesses is a prompt with extra steps.
Install the package and run the lint command over your documentation directory. Add the GitHub Action to upload SARIF, and each finding appears in code scanning on the line that caused it. Point your agents at the skill adapter or the MCP server, so the same rules apply while the text is written, not after review.