| name | codex-instructions-creator |
| description | Create or revise layered AGENTS.md instructions with clear scope, verified repository facts, native Codex boundaries, and concise operating rules. Use when adding project, directory, or global agent guidance. |
Codex Instructions Creator
Resolve every ../../scripts and ../../references path relative to this SKILL.md directory, never the caller working directory.
Write the narrowest instruction layer that helps Codex act correctly in its directory scope without duplicating lower-level truth.
Workflow
-
Determine the file's scope, parent instruction chain, target audience, and facts that genuinely need durable guidance.
-
Read ../../references/codex-artifact-contracts.md for discovery, precedence, fallback, and size rules.
-
Inspect options and generate:
python3 ../../scripts/create_codex_artifact.py instructions --help
python3 ../../scripts/create_codex_artifact.py instructions --output . --name repository-instructions --description "Repository instructions for Codex"
-
Replace scaffold text with verified purpose, boundaries, workflows, safety constraints, and exact verification commands.
-
Keep code and config as source of truth; reference canonical files instead of copying volatile versions or pins.
-
Use skills and plugins for detailed workflows. Do not create slash-command instructions or collapse another CLI's first-class file into an import.
-
Validate:
python3 ../../scripts/check_codex_artifact.py instructions AGENTS.md
-
Review the full parent-to-child chain for contradictions, duplication, and stale paths.
Quality bar
- Every rule is actionable, scoped, and based on repository truth.
- More specific files refine parent guidance without silently reversing safety boundaries.
- Commands are executable from the stated directory.
- No secrets, chat transcripts, speculation, or runtime-local state are stored.