CodeSmith — Mentor Package
Specialty: Full-stack dev automation — CI/CD, code review, GitHub workflows, ACP dispatch
Version: 1.0.0
For: OpenClaw agents that do real coding work — implementing features, managing repos, dispatching to sub-agents
What This Package Contains
This is a mentor package consumed by the claw-mentor-mentee skill. It teaches a subscriber's agent how to operate as a serious coding-focused setup.
| File | What It Teaches |
|---|---|
AGENTS.md | Annotated configuration — 17 annotation blocks explaining the why behind every non-obvious decision |
working-patterns.md | Daily coding rhythm, ACP dispatch patterns, trust progression, 5 real failure stories |
skills.md | Tier 1/2/3 skill stack + skills explicitly NOT installed (with reasons) |
cron-patterns.json | 5 cron jobs with adoption guide — add one at a time |
privacy-notes.md | Explicit read/write/network access tables |
setup-guide.md | Step-by-step onboarding with Relationship Adoption Timeline |
CLAW_MENTOR.md | Full package manifest with risk assessment and compatibility notes |
Who This Is For
A developer who uses OpenClaw as a coding partner and wants that partner to operate with more autonomy, better judgment, and fewer surprises. Assumes:
- GitHub account with
ghCLI configured - A hosting service (Vercel or equivalent)
- ACP enabled for sub-agent dispatch
- Comfort with some agent autonomy once trust is established
Not for setups where the human reviews every single change, or for purely non-technical workflows.
How to Apply
This package is applied automatically by the claw-mentor-mentee skill (v2.1.0+) during your scheduled ingestion cycle.
Manual review recommended before any cron jobs are enabled — see setup-guide.md for the one-at-a-time adoption timeline.
About CodeSmith
Built from real production work: implementing API endpoints, debugging deployment pipelines, managing GitHub workflows, dispatching coding sub-agents for implementation tasks, and learning the hard lessons that only come from things actually breaking in production.
The failure stories in working-patterns.md are real. The cron timing is what actually ran. The trust progression is how it actually builds. That's what makes it useful.