| name | setup-mrbaeksang-skills |
| description | Configure the mrbaeksang rails in a repo — record its deployment level and confirm the enforcing hooks are active. Run this once after installing the plugin. Use when the user runs /setup-mrbaeksang-skills or asks to set up / configure the rails. |
| disable-model-invocation | true |
Setup mrbaeksang skills
The rails calibrate everything off one number: the repo's deployment level. This skill
records it and confirms the enforcing hooks are live. Run it once per repo.
Steps
-
Explain, then ask the deployment level. It's the master dial — the user may not know
the term. Levels:
- L0 — local / prototype. Just making it work. Rails off; lightest.
- L1 — staging / internal. Real but limited users. Research full, smoke-gate auto, hooks warn.
- L2 — production. External/paying users, real responsibility. Full enforcement; hooks block.
Recommend L2 unless the user says it's a throwaway. Never assume — always ask.
-
Write docs/agents/deployment-level.md with two machine-readable markers, each on its own line:
<!-- mrbaeksang:current=L1 -->
<!-- mrbaeksang:target=L2 -->
current = what hooks enforce now (block/warn/off); target = what you design toward (seams).
Then the human-readable dial table. The markers are the single source of truth hooks grep.
-
Seed .claude/usage-rules.tsv from the plugin's rules/starter-usage-rules.tsv (common deps'
official-usage rules). The generic usage-guard hook reads this file; /onboard-dependency
appends a line per new dependency. The hook code never changes — only this data grows.
-
Write the ## Engineering rails section into CLAUDE.md (or AGENTS.md if that's the repo's
file) so a fresh agent reads the conventions before hooks ever fire:
## Engineering rails (mrbaeksang)
- Deployment level: docs/agents/deployment-level.md (current/target). Build for target, run at current.
- External dependency? Onboard first (/onboard-dependency): research official usage grounded in the
installed source → vendor docs/official/<dep>/CONSIDERATIONS.md → append usage rules. Don't use a
library you haven't onboarded.
- No dodge: never ship "overkill / blocked-on-credential / not sure" — check level / use a test
double or local instance / research first.
- Mechanism lives in code: ERD = the migration, API = generated OpenAPI. Don't hand-author
erd.md/api.md. Decisions → docs/adr/, terms → CONTEXT.md.
- Hooks (.claude, level-calibrated): onboard-dependency-guard, doc-discipline-guard, usage-guard.
-
Confirm rails are active. /reload-plugins should report the mrbaeksang-rails hooks.
Smoke-test: add an un-onboarded dependency to a manifest (blocked at L2 / warned at L1), and ensure
docs/official/INDEX.md exists for vendored sources.
Which skills now read this
/onboard-dependency — gates new dependencies; rigor + block/warn/off scales with the level
/smoke-gate — strictness of the "look at the real output" gate scales with the level
/due-diligence, /spike-first — mandatory at L2, optional at L0
Done
Tell the user the level is recorded, the rails are calibrated, and they can change it anytime by
editing the mrbaeksang:level= marker — re-running this skill is only needed to reconfigure.