apps

name	apps
description	이 스킬은 사용자가 팀에 등록된 axhub 앱 목록을 보거나 명시적인 앱 관리 작업을 요청할 때 사용해요. 다음 표현에서 활성화: "내 앱", "내 앱 보여줘", "내 앱 봐", "등록된 앱", "등록된 앱 봐", "앱 등록", "앱 리스트", "앱 목록", "앱 목록 보여주세요", "앱 목록 봐", "앱 뭐", "앱 뭐 있어", "앱 보여", "앱 봐", "앱 삭제", "앱 생성", "앱 슬러그", "앱 슬러그 봐", "앱 제거", "앱 지워", "앱 id", "앱 ID 봐", "어떤 앱", "어떤 앱 있어", "어떤 앱이 있나요", "우리 앱", "우리 앱 봐", "운영 중인 앱", "운영 중인 앱 뭐 있어", "운영 중인 앱 보여주세요", "제 앱", "제 앱들 보여주세요", "회사 앱", "회사 앱 뭐 있어", "app catalog", "app list", "apps", "apps create", "apps delete", "apps rm", "available apps", "list apps", "my apps", "which apps", 또는 앱 카탈로그/관리 흐름. 현재 팀 scope 으로 출력 필터링하고 생성/수정/delete 작업은 승인 토큰을 요구해요.
examples	[{"utterance":"내 앱 목록 보여줘","intent":"list axhub apps"},{"utterance":"이 앱 삭제해","intent":"list axhub apps"},{"utterance":"list my apps","intent":"list axhub apps"},{"utterance":"list apps","intent":"list axhub apps"},{"utterance":"내 앱 봐","intent":"list axhub apps"}]
multi-step	false
needs-preflight	true
allows-dependency-execution	false

Apps Management (team-scoped; mutations consent-gated)

Show registered axhub apps for the current team. Listing/details are read-only; create, update, and delete paths require an AskUserQuestion preview plus HMAC consent token before any mutation command.

Pre-execute preflight context (Phase 17 US-1706 — !command injection):

!`${CLAUDE_PLUGIN_ROOT}/bin/axhub-helpers preflight --json`

이 줄은 Claude Code SKILL preprocessing 으로 워크플로 시작 전에 실행돼요. 출력 (auth_status, current_app, current_env) 이 컨텍스트에 자동 주입돼서 별도 auth/profile 호출이 줄어요.

Workflow

To list apps:

1. Pre-flight (lightweight). Confirm auth before the list call:

   ${CLAUDE_PLUGIN_ROOT}/bin/axhub-helpers preflight --json
   

   On auth_ok: false, halt and route to ../deploy/references/error-empathy-catalog.md ("exit 65"). Suggest the auth skill via "다시 로그인해줘".

2. Fetch apps:

   axhub apps list --json
   

3. Filter to current team scope. Drop entries whose team_id does not match $AXHUB_TEAM_ID (or the team derived from axhub auth status --json). Do NOT dump cross-team apps to the user — they are surface noise that breaks the F4 privacy guarantee.

4. Render top 10 in Korean. Format as a numbered list with slug (id=N) — <status> per row:

   현재 팀 앱 10개 (전체 N개):
     ① paydrop (id=42) — production: succeeded (12분 전)
     ② paydrop-staging (id=43) — staging: succeeded (1시간 전)
     ③ checkout-svc (id=44) — production: failed (어제)
     ...
   

Non-interactive AskUserQuestion guard (D1): 이 SKILL 의 모든 AskUserQuestion 호출은 대화형 모드를 가정해요. if ! [ -t 1 ] || [ -n "$CI" ] || [ -n "$CLAUDE_NON_INTERACTIVE" ] 인 subprocess (claude -p, CI, headless) 에서는 AskUserQuestion 호출을 건너뛰고 안전한 기본값으로 진행해요. 기본값은 tests/fixtures/ask-defaults/registry.json 참조 — expansion → skip (top 10 으로 충분), delete confirmation → abort (삭제 안 함).

5. Offer expansion. If the filtered list exceeds 10, surface AskUserQuestion:

   {
     "question": "앱이 더 있어요. 전체 목록 볼래요?",
     "header": "전체 보기",
     "options": [
       {"label": "네, 전체 보기", "value": "show_all", "description": "현재 팀의 모든 앱"},
       {"label": "지금은 그대로", "value": "skip", "description": "상위 10개로 충분"},
       {"label": "검색 (slug 입력)", "value": "search", "description": "특정 slug 검색"}
     ]
   }
   

6. On validation.app_list_truncated (>100 apps server-side): route to ../deploy/references/error-empathy-catalog.md ("exit 64 + validation.app_list_truncated"); ask user to provide a numeric --app <id> directly.

7. On non-zero exit, route to ../deploy/references/error-empathy-catalog.md by exit code (65 / 67 / 68 / 1). Read paths may auto-retry once on exit 1.

v0.2.0 command coverage polish

Use these paths only when the user intent is explicit. Listing remains the default.

apps create

1. Preview the source file or interactive intent first.

2. Mint consent with stdin JSON:

   printf '%s\n' "$CONSENT_BINDING_JSON" | ${CLAUDE_PLUGIN_ROOT}/bin/axhub-helpers consent-mint
   

   Required binding fields: action=apps_create, context={source}.

3. Run one of the current CLI contracts:

   axhub apps create --from-file apphub.yaml --yes --json
   axhub apps create --interactive --json
   

apps get

Read-only details use:

axhub apps get "$APP" --json

apps update

Preview each key=value field, then mint action=apps_update with top-level app_id and context={slug,field} before:

axhub apps update "$APP" --field "$FIELD" --json

apps delete

Deletion is consent-gated. Do not run axhub apps delete ... --dry-run --json before approval; the hook parser currently treats every apps delete shape as destructive, including dry-run.

1. Build the preview only from read-only data:

   axhub apps list --json
   axhub apps get "$COMMAND_TARGET" --json
   

2. Define one target and keep it unchanged through the whole flow:

   COMMAND_TARGET="$APP"
   

   Prefer the exact slug the user typed or selected. If the user selected a numeric id instead, use that exact numeric id. The preview may show both slug and numeric id, but consent-bound fields use only COMMAND_TARGET.

3. Ask for exact confirmation before minting a token:

   {
     "question": "앱을 삭제할까요?",
     "header": "앱 삭제",
     "options": [
       {"label": "삭제", "value": "delete", "description": "표시한 COMMAND_TARGET 앱을 삭제해요."},
       {"label": "취소", "value": "abort", "description": "삭제하지 않아요."}
     ]
   }
   

   In non-interactive mode, use the registry safe default abort and stop.

4. Mint consent with the literal command-target invariant. For apps_delete, context.slug is the parser field name and may contain a numeric id when COMMAND_TARGET is numeric.

   # Binding shape: {"action":"apps_delete","app_id":"$COMMAND_TARGET","context":{"slug":"$COMMAND_TARGET"}}
   CONSENT_BINDING_JSON=$(jq -nc \
     --arg target "$COMMAND_TARGET" \
     '{tool_call_id:"pending",action:"apps_delete",app_id:$target,profile:"",branch:"",commit_sha:"",context:{slug:$target}}')
   printf '%s\n' "$CONSENT_BINDING_JSON" | ${CLAUDE_PLUGIN_ROOT}/bin/axhub-helpers consent-mint
   

5. Run exactly one delete command using the same target string:

   axhub apps delete "$COMMAND_TARGET" --yes --json
   

apps open delegation

If the user wants to open a live app or dashboard, route to ../open/SKILL.md instead of using axhub apps open.

NEVER

• NEVER list cross-team apps without explicit user opt-in (F4 privacy guarantee).
• NEVER dump >10 rows in the first response (overwhelms vibe coders).
• NEVER drop --json (parsing depends on it).
• NEVER cache app_id locally for deploy mutation paths — the deploy skill must live-resolve.
• NEVER remint apps delete consent with numeric id when the command will use a slug, or with slug when the command will use a numeric id. Keep COMMAND_TARGET identical.
• NEVER echo internal endpoint URLs of cross-team apps even if visible in stdout.

Additional Resources

For Korean trigger lexicon (apps intent): ../deploy/references/nl-lexicon.md. For 4-part Korean exit templates: ../deploy/references/error-empathy-catalog.md. For privacy filter rules (cross-team scope, NFKC normalize): see the redact subcommand in axhub-helpers and PLAN §16.17.