메뉴
AUTO-INVOKE when user asks about the agent's version or update status. Triggers: version / 版本 / 版本信息 / update status / 更新状态 / self check / 自检. Reports installed version, index latest, pinned state, and update-loop health evidence.
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
AUTO-INVOKE for fast technical scans and market briefs. Triggers: scan TICKER (or 扫 TICKER, quick TICKER) for single/multi quick analysis; portfolio health (or portfolio-health, 持仓健康) for batch portfolio scan; market brief HK/US (or brief HK/US, 市场简报 HK/US) for index+sector overview. Output is a DSA-style 决策看板 / 市场简报 emitted on stdout for you to deliver.
AUTO-INVOKE for structured adversarial review when the user wants to stress-test a plan, decision, thesis, or article angle. Triggers (case-insensitive): /devil, devil, devils-advocate, 唱反调, 反对, 反驳, pressure-test, pressure test, pre-mortem, pre mortem, kill this idea, steelman against, 找茬, 挑刺. The skill produces a strict 3-section adversarial output. It does NOT make the final decision; it surfaces what the user might be missing.
Monthly portfolio allocation review — load rules + latest valuation, compute bucket weights vs targets, identify rebalance candidates, enforce gold/cash/single-name caps, save JSON+TXT, and deliver a concise summary.
Policy-constrained Portfolio Manager (PM) decision workflow — compute investable/bucket weights, diagnose drifts and breaches, and emit three-layer recommendations with audit + chat rendering.
Detect and reject fabricated portfolio reports with fake holdings (AAPL, TSLA, GOOGL) vs real anchors (NVDA, NVDA, MSFT, 0700.HK, 0700.HK, Japanese/Singaporean funds)
Deterministic portfolio valuation and chat-friendly push using Yahoo multi-endpoint fallbacks (v7 quote → v8 chart → v7 spark), static valuation for MMF/StructNote/CODE items, at-cost valuation for HSBC 2028 bond, FX overrides, MMF safety allocation, ringfenced obligations, liabilities deduction, separate lots for duplicate tickers (e.g., 0700.HK A/B), JSON snapshotting, and cron scheduling at SGT 18:00 with manual /888 trigger.
| name | version-info |
| description | AUTO-INVOKE when user asks about the agent's version or update status. Triggers: version / 版本 / 版本信息 / update status / 更新状态 / self check / 自检. Reports installed version, index latest, pinned state, and update-loop health evidence. |
When the user asks any of: version, 版本, 版本信息, 你是哪个版本, update status, 更新状态, 自动更新正常吗, self check, 自检 — you MUST run the script below and deliver its output. Do NOT answer version questions from memory: your session's system prompt is a snapshot from session creation and may be OLDER than what is on disk.
Run via execute_code:
import subprocess, sys
result = subprocess.run(
[sys.executable, "${CLAUDE_PLUGIN_ROOT}/skills/ops/version-info/scripts/version_info.py"],
capture_output=True, text=True, timeout=60,
)
print(result.stdout)
sys.stderr.write(result.stderr)
The script is pure stdlib (no venv required). It needs network access to fetch the public index; if that fails it still reports local info and says so.
The script prints the full report between __REPORT_START__ and __REPORT_END__, followed by a DONE:/FAILED: status line. Reply with the text between the markers verbatim — no paraphrasing, no summarizing. You MAY append a short interpretation after the report (e.g. what the verdict means for the user), clearly separated from it.
.agentspec-content/active.json（E2B 内容更新器的 active release，最权威）→ .agentspec.json install marker（含 pinned）→ agent.json / agent.lock.json fallback；都没有则明确说这是 dev 环境或安装未完成。profile 目录自动推导：PROFILE_DIR env → HERMES_HOME env → 从脚本自身路径向上找 → ~/.aisa/agents/aisa-cio。https://raw.githubusercontent.com/AIsa-team/agent-index/main/index.json 比较。content-update.log（$HOME 或 profile 下）尾部各轮 status + 最后活动时间，直接判断循环是否活着。content-update.log 报错行。pinned=true → 显式钉住，符合预期。