원클릭으로
jdi-create
Creates new JDI agent or skill via validated question loop and automatic integration.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Creates new JDI agent or skill via validated question loop and automatic integration.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Registers a new phase in ROADMAP.md. Slug-as-ID — multi-developer safe. Validates slug uniqueness and shape. Append at end (default), or position via --before/--after. Atomic commit.
Adopt mode for brownfield projects. Scans existing repo (manifests, layout, git, docs), infers stack/code-design, confirms with user, generates PROJECT.md + ROADMAP.md with adopted=true flag. Replaces /jdi-new for projects with code already written.
Removes a phase from ROADMAP.md. Accepts slug OR position. Refuses to remove done/current/past phases. Archives any existing phase artifacts. Atomic commit.
Upfront pre-roadmap research. Reads user idea, asks key questions, captures project-wide Definition of Done baseline, researches stack/domain, generates initial PROJECT.md + ROADMAP.md. Single agent instead of multiple parallel researchers to save tokens.
Prints a compact summary of where the project is — current phase, what the last action did, and the exact next command to run. --stats adds outcome metrics derived from artifacts (first-pass rate, loop iterations, lead time). Read-only. No agent invoked.
Adaptive question loop to capture locked decisions before planning the phase. Accepts slug or position.
| name | jdi-create |
| description | Creates new JDI agent or skill via validated question loop and automatic integration. |
| argument_hint | [optional short description] |
| runtime_intent | {"invokes_agent":"jdi-architect"} |
| runtime_overrides | {"claude":{"allowed-tools":["Read","Write","Edit","Bash","Grep","Glob","AskUserQuestion","Agent"]},"copilot":{"tools":["read","write","edit","terminal"]},"antigravity":{"triggers":["/jdi-create","create new agent","create new skill","extend jdi"]}} |
Examples:
/jdi-create/jdi-create "specialist for Rust with cargo + clippy"/jdi-create "reviewer focused on a11y for UI"/jdi-create "skill with EF Core 9 conventions"
Contributor-only guard — this command writes agents/skills into core/ of
the jdi-cli SOURCE repo. A consumer project that happens to have its own
core/ folder must never pass this gate:
grep -q '"name": "jdi-cli"' package.json 2>/dev/null || {
echo "/jdi-create runs only inside the jdi-cli source repo (it writes to core/)."
echo "Consumer projects extend JDI via per-project specialists — see EXTENSION.md."
exit 1
}
test -f core/agents/jdi-architect.md || { echo "core/ incomplete — run from the jdi-cli repo root."; exit 1; }
# Clean core/ required (avoid mixing generated files with local edits)
DIRTY_CORE=$(git status --porcelain core/ 2>/dev/null || true)
[ -z "$DIRTY_CORE" ] || { echo "core/ has uncommitted changes. Commit or stash first."; exit 1; }
Invoke jdi-architect:
Wait. Architect runs 12 steps (see core/agents/jdi-architect.md).
Architect returns 1 of 3 statuses:
If created:
jdi-{name} ({type}) ok. Audit: R-{date}-{name}. Commit: {sha}.
Invoke: {runtime instructions}
- pre: running inside the jdi-cli source repo (`package.json name == jdi-cli` + `core/agents/jdi-architect.md` present) + `core/` clean
- post: agent/skill created + integration points updated + build+install done + atomic commit
- Not the jdi-cli source repo -> abort with pointer to EXTENSION.md (consumer path)
- Working tree dirty in `core/` -> ask to commit or stash first
- User cancelled -> exit with no side effects
- Build failed -> do not install, show build error, keep core/ updated for manual retry