Using ASD-STE100 Principles to Write Clearer Codex Prompts

I’ve used an ASD-STE100-inspired skill in Codex to improve my prompts. A few of its habits fit coding-agent instructions well: use one name for one thing, make actions explicit, and keep instructions short.

ASD-STE100 is a controlled language for technical documentation. It has writing rules and a controlled dictionary. I’m borrowing those habits, not claiming that my prompts meet the standard. You can request a copy from the official downloads page.

Stop making the agent guess

Say a prompt calls the same value a “query,” a “filter,” and a “search term.” Does Codex treat those as one thing or three? Call it a query throughout. “Add a test for an empty query” gives Codex an action. “Address empty-query coverage” leaves more to interpret.

Here is a fictional recipe app task. The first prompt sounds reasonable until you try to implement it:

1Make the recipe search better.

“Better” could mean almost anything. Once the human chooses the behavior, the prompt can say what to build:

1Update the recipe list search.
2Match recipe titles without case sensitivity.
3Show all recipes when the query is empty.
4Show an empty-state message when no titles match.
5Use the existing UI conventions.
6Do not add dependencies.
7Add tests for case matching, empty queries, and no matches.
8Run the recipe search tests. Report the results.

None of those requirements came from the vague line. The human had to choose the search scope, matching rule, empty-query behavior, no-match behavior, dependency limit, and tests. A tidy rewrite of “make search better” would still be missing those decisions.

Try it on your own prompts

If you installed $asd-ste100 in your Codex setup, invoke it by name. Codex supports explicit skill invocation. This example requires a skill with that name in your setup.

1$asd-ste100 Rewrite these Codex instructions for clarity.
2Preserve their meaning, conditions, and uncertainty.
3Use consistent terms and explicit actions.
4Do not add requirements or facts.
5Ask me when the meaning is unclear.
6
7Instructions:
8[paste your instructions]

If you don’t use that skill, paste this instead:

 1Rewrite these instructions for a coding agent.
 2Use active voice.
 3Use consistent names for the same things and actions.
 4Put one action in each instruction.
 5Keep instructions short.
 6Preserve meaning, conditions, and uncertainty.
 7Do not add requirements.
 8Ask me when the meaning is unclear.
 9
10Instructions:
11[paste your instructions]

Read the rewrite before using it. If “may” becomes “will,” the meaning changed. If the task still lacks a requirement, decide it yourself or ask a question. The missing behavior remains your call.