jaewhi-park
UserAn autonomous dev team in one folder of config files — six agents that plan, build, verify, review and ship. Works in Claude Code and opencode.
Categories
Indexed Skills (29)
hire
Hiring — design per-role model assignments and an operating profile from available models, budget and project character; applied on the budget answer (from /kickoff, /assess or this command) and shown as a roster — the CEO changes a line by saying so. Right after /kickoff or /assess, when the budget changes, or when the team is too slow or expensive
ship
Ship — full verification, docs, risk label, automatic merge-policy resolution (local / manual / auto-low-risk), then PR creation or local merge
assess
Legacy assessment from the approved brief — inventory, risks, rebuild strategy (ADR), parity harness plan for the first seam. Run /brainstorm first; rebuild only after the harness exists
build
Implement an approved plan step by step (the implementer proves each step red → green; the verifier gives the full verdict at the end)
claude-md
How to write and update the repository's CLAUDE.md (commands, architecture map, conventions, Definition of Done, forbidden actions) and its template. Load at kickoff, legacy assessment, when a retro adds a rule, or when CLAUDE.md is missing or its commands no longer match reality.
integrate
Integrate finished branches one at a time into an integration branch cut from main — per-branch 4-lens review → merge → conflicts resolved by the implementer → verifier PASS → next (merge queue); /ship then ships that branch
kickoff
New project kickoff from the approved brief — one question (stack and budget), then charter, stack ADR, skeleton, rules file, the product spec (docs/specs/), backlog, first plan and the hire, all by the team. Run /brainstorm first
resume
Resume — reconcile the status board, the plan, worktrees and git, then continue from the next action. The lead runs it itself at the first turn of a session that injected a board and after any interruption
review
Review a diff — the test-file floor, then 2 lenses (correctness/security) or 4 depending on risk, in parallel → fix loop
brainstorm
Product brief (기획서) — brainstorm freely with the lead, or bring a document you wrote, and get docs/BRIEF.md compiled, gap-checked, critiqued and approved. Run before /kickoff or /assess; run again to revise the brief (pivot)
design
What a user-facing change must have to be finished — screens and their four states, layout and hierarchy, copy, responsiveness, keyboard and focus, consistency with the design foundation — and how it is proven (the `shows:` lines of a UI step, screenshots from the rules file's screenshots command). Loaded by team-planner when a plan touches a UI path, by team-implementer and team-builder for a UI step, and by team-reviewer for the ux lens; at /kickoff for the design foundation.
instruction
What the lead does with what the CEO says — the triage of every message (question / instruction / remark / explanation in progress / go / stop) and the procedure for an instruction (classify → write it down → impact → sort the plans → continue). Loaded by the lead at every CEO message that is not a plain "계속"; never invoked by the CEO.
security
The security baseline a product starts with, the threat-model lines a plan carries when it touches a Risk path, the negative-case `proves:` lines that make them testable, and the checklist the reviewer's security lens works through. Loaded at /kickoff for the baseline, by team-planner for a plan on a Risk path, by team-reviewer for the security lens, and by team-critic when it checks either.
spec
Structure, file template and revision rules of the product spec — docs/specs/F<nn>-<slug>.md, one file per feature, with docs/SPEC.md as the index. Written by team-planner from the brief at /kickoff, never through CEO rounds; extended when a feature arrives later; revised through /plan's `수정:` path. read by team-planner before writing or revising it and by team-critic when reviewing it.
backlog
Product function — turn ideas and requests into backlog items with completion criteria and prioritize them (docs/BACKLOG.md, optionally GitHub Issues)
handoff
Wrap up a session that stops mid-flight — write and commit the status board (docs/STATUS.md), commit WIP, list pending decisions. A finished /ship or /plan already leaves board and commits in place; use this when ending mid-step, after repeated compaction, or when the team spins
lang
Set the team's working language — writes the "## Language" section of CLAUDE.md via script and switches responses immediately. Use when the team answers in the wrong language or you want to change it
parallel
Run several approved, independent plans concurrently — one team-builder per plan in its own isolated worktree. Integrate afterwards with /integrate
policy
Show/set the merge policy — current automatic verdict (local / manual / auto-low-risk) with its reason; with an argument, record an override in the rules file
recruit
Recruit a new role — when no existing agent fits (a different permission boundary or a separate judgment is needed), propose one to the CEO together with its model and create it on approval. The lead suggests it from /plan or /retro, never mid-build
release
Release — version decision, finalized CHANGELOG, release notes, tag commands. Tag push is the CEO's
retro
Retrospective — feed the causes of rework, failures, CEO intervention and incidents back into rules, skills or guardrails
roster
Team roster — table of each agent's model, permissions and role, plus how to change budget profile or individual models (read-only)
run
Default path — take an approved plan through build → review → ship. Stops only at escalation, fix-loop overflow and a risk:high merge (the CEO's)
brief
Template, slots and writing rules for the product brief (docs/BRIEF.md, the 기획서) — product level, in the CEO's words, never injected. Loaded by /brainstorm; read by team-planner when writing, normalizing or revising the brief, and by /kickoff and /assess when they derive the charter from it.
agents-md
How to write and update the repository's AGENTS.md (commands, architecture map, conventions, Definition of Done, forbidden actions) and its template. Load at kickoff, legacy assessment, when a retro adds a rule, or when AGENTS.md is missing or its commands no longer match reality.
charter
Slot list and template for the project charter (docs/CHARTER.md), derived from the product brief. Load at /kickoff and /assess when the charter is written, or whenever goals, scope or non-goals are unclear or drifting.
legacy-assessment
Checklist for assessing a legacy codebase: inventory, risk register, seam identification, rebuild-strategy comparison and the docs/ASSESSMENT.md format. Load when starting a rebuild, modernization or large refactor of an existing system, or before changing an unfamiliar codebase substantially.
parity-harness
Procedure for pinning current behaviour before a legacy rebuild with a characterization-test / golden-output harness, and the docs/PARITY.md format. Load as the first step of any replacement, rewrite or migration of legacy code, and use it for regression verdicts at every rebuild step.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.