Source-reviewed skill for agentic software development with Codex, GitHub, OpenClaw, and ClawHub.
The skill is intentionally instruction-only:
- no scripts
- no installers
- no credential readers
- no background daemons
It teaches a Karpathy-style operating loop: state the goal, inspect the repo, assign explicit roles when needed, make scoped changes, verify, review for leaks and regressions, update durable memory, then publish only when the public surface is clean.
SKILL.md: runtime skill instructions and metadata.agents/openai.yaml: Codex UI metadata.references/source-review.md: source analysis behind the workflow.references/comparison-matrix.md: source-by-source comparison and gap analysis.references/system-design.md: task, memory, report, role, and isolation templates.references/example-run.md: real-run eval acceptance shape.references/publish-checklist.md: GitHub and ClawHub release checklist.
See SECURITY.md for responsible disclosure and the release bleed gate.
python3 -m antirot.cli lint SKILL.md --strict
clawhub publish . --version 0.3.1
clawhub inspect agentic-codex-dev --filesMIT-0.