Integration guide language review

The Codex guide uses ASD-STE100 Issue 9, 2025-01-15 as its writing reference. This record covers the new integration guide and its technical terms. It does not claim independent certification or approval by ASD.

Language scope

The inspection used Part 1 for words, verb forms, active voice, sentences, procedures, descriptions, and word counts. It also used the Part 2 dictionary to examine word meanings and parts of speech. The inspection replaced ordinary words that the dictionary does not approve. Examples include “requires,” “otherwise,” “already,” and “reason.” The guide uses “check” as a noun and “examine” as an instruction verb.

The guide defines its technical terms before the procedures. Computer terms use rule 1.5, category 19. Commands, result states, and field names keep the source spelling under category 10. Technical verbs identify computer processes only. The term list does not give permission to use ordinary unapproved wording as technical language.

The inspection examined each prose sentence for its word count, instruction count, verb form, and voice. The guide has one instruction per sentence and active voice. Description paragraphs have one topic and a maximum of six sentences. The word-count check treats each code quotation as one item. It does not count code blocks as prose sentences.

Source and example checks

The CLI and MCP descriptions agree with the product source files cf/nerudo-cli/src/main.rs and cf/nerudo-cli/src/mcp.rs. The official Codex MCP documentation gives the configuration field names and timeout meanings. Both product source links gave HTTP 200 during the inspection. The new guide URL cannot be available before repository approval and publication.

The local link check passed for all Markdown links and 14 navigation pages. git diff --check passed in both repositories. The guide sentence check found a maximum of 19 words per instruction and 21 words per description sentence.

The Python launcher example compiled. Read-only live checks also passed through the extracted example with operator-supplied agent credentials. Its HTTP record mode gave the expected landed revision and full input commit ID. Its MCP mode gave a LANDED receipt for the same change revision. The checks did not submit a change through the example. The temporary private credential file was deleted after the checks. The launcher uses owner-only credential files, a fixed HTTPS origin, no HTTP redirects, and a named user agent. A public HTTP check showed that the default Python user agent received 403. The example user agent received 200 for the same public repository route.

The procedure distinguishes the AGENT bearer token from a personal developer credential. It uses the full input commit ID and an explicit revision number for receipt checks. It also compares the receipt with Git and the landing log. The native tools give text results. The HTTP record gives JSON fields directly. The procedure stops at a required human seal.

Limits

The dictionary and grammar inspection was manual, with a separate word-count check. The count check cannot prove vocabulary or grammar compliance. The inspection is not an independent language assessment. Future source or interface changes can make command examples incorrect. Maintainers must examine the examples when these interfaces change.