| name | codex-rule-creator |
| description | Create or revise native Codex execpolicy .rules files with narrow command prefixes, explicit decisions, and executable positive and negative examples. Use when governing shell-command approval behavior. |
Codex Rule Creator
Resolve every ../../scripts and ../../references path relative to this SKILL.md directory, never the caller working directory.
Create a minimal execpolicy rule that classifies one command family without granting unintended variants.
Workflow
-
Define the exact command tokens, desired decision, trust rationale, platforms, and both allowed and disallowed examples.
-
Read ../../references/codex-artifact-contracts.md for the current rules language, locations, and evaluator behavior.
-
Inspect options and generate:
python3 ../../scripts/create_codex_artifact.py rule --help
python3 ../../scripts/create_codex_artifact.py rule --output .codex/rules --name safe-shell --description "Execpolicy for a bounded command family" --prefix git status
-
Use token-aware prefixes, not shell substrings. Prefer multiple narrow rules over one broad pattern.
-
Include justification and match examples where supported. Treat shell wrappers, flags, paths, and subcommands as separate attack surfaces.
-
Do not encode secrets, environment-specific home paths, or blanket allow rules.
-
Validate:
python3 ../../scripts/check_codex_artifact.py rule .codex/rules/safe-shell.rules
-
Run the native rules evaluator against positive, negative, wrapper, injection, and near-miss cases when available.
Quality bar
- Every allow decision is bounded to the intended argv shape.
- Negative examples cover sibling subcommands, added shell operators, wrappers, and path escapes.
- Overlap between rules has a deliberate outcome.
- The file is portable, documented, and fails closed on ambiguous input.
