Portfolio Risk Review

Single-pass portfolio review. For multi-role analyst framework, use portfolio-swarm-review instead.

Required reads

In this order:

1. /root/.openclaw/workspace/MEMORY.md — rules, traps, user preferences
2. /root/.openclaw/workspace/portfolio.json — authoritative holdings
3. /root/.openclaw/workspace/memory/theses/*.json — canonical thesis state when present
4. ../daily-deep-brief/references/technical-playbooks.md — read before any add / average-down plan
5. /root/.openclaw/workspace/TOOLS.md — data chain reference if anything fails

portfolio.json is the only active/exited list: shares > 0 is held, shares == 0 is exited. There is no hand-maintained summary any more — the one that used to sit here went 3.5 months without a sync and ended up naming seven held positions as exited (#1067).

The thesis registry is read-only during a portfolio review. Missing files mean unknown, not permission to reconstruct a baseline from prior prose. Price moves may change valuation but cannot by themselves change business, moat, or management.

Fresh data rule

Always refresh quotes before judging the book. Use the workspace scripts.

# US: 7-route fallback, RSI/MA/news/signal, writes back to portfolio.json
/root/.local/bin/clawock analyze-us
/root/.local/bin/clawock analyze-us --no-news    # skip news

# HK: Tencent primary + Eastmoney full-batch independent cross-check/fallback → stooq → yfinance
/root/.local/bin/clawock analyze-hk
/root/.local/bin/clawock analyze-hk --no-news

Use the returned quote coverage/source warnings to mark stale or unverified holdings; failure of an unused fallback alone does not invalidate a successful quote. Special trap: 00100 only has Tencent — Tencent down means 00100 is stale.

KR linkage: 07709/07747 are frequently at shares == 0, and SKHY (SK Hynix ADR) is an independent line — read the KR names off portfolio.json (shares > 0) rather than assuming any part of the chain is held or dead.

Holdings bucketing

Do not hardcode tickers here — names rotate. Pull the active list from portfolio.json (shares > 0), then bucket by category each run:

The framing is stable; the contents drift. Verify each session against portfolio.json.

Four-lens analysis

Lens A — PnL and position quality

For each active holding:

• gain/loss in $ and %
• distance to breakeven
• thesis status: intact / weakening / broken
• holdable / trim / T-only / cut

Lens B — Cross-market linkage

US side (typically):

• NASDAQ / 纳指夜盘 tone — drives next-day HK tech open
• Specific theme threads (e.g. stablecoin reg for CRCL, space/defense for RKLB)

HK side:

• 恒科指数方向 — primary driver for 03032 / 03033 / 07226
• 南向资金当日净流向 (web search when material)
• Sector policy: 风电 for 02208, AI capex for 00100

Note explicit "supportive / neutral / weak" tag for each chain.

Lens C — Concentration and drawdown risk

• Largest $ loss contributors right now
• Leverage decay risk (any 2x/3x ETF held > 5 trading days?)
• Correlation clusters (e.g. multiple 杠杆 ETF + single high-beta name = one bet)
• One-bad-day worst case for the book

Lens D — Action priority

Sort positions into:

• Hold and watch — thesis intact
• Trim on rebound — thesis weakening but not broken; wait for strength
• T-only — don't add, exit on bounces, no overnight conviction
• Add only on trigger — define the trigger explicitly (price, MA cross, earnings, policy)

Keep core_position, tactical_entry, and risk_rebalance separate. A concentrated non-leveraged core may remain held while a tactical add is proposed. Registered setups are suggestions: a self-defined entry states a hypothesis and invalidation price, and policy-tranche departures are recorded. Loss alone is not evidence of a good entry. For HK, state board lots and use whole-lot multiples; missing lot_size blocks an add. For US, use integer shares under the current ledger contract.

Output structure

Portfolio snapshot

• 总 US PnL / 总 HK PnL
• 最大盈利位 / 最大亏损位
• 主要风险源 (one line)

Position-by-position

Table format:

Cross-market read

• US side: tone + key threads
• HK side: 恒科方向 + 南向 + sector policy notes

Top 3 risks (ranked)

Highest to lowest, each with concrete cause.

Action plan

Four buckets from Lens D, with concrete triggers/levels where applicable.

Style rules

• Direct, practical, no academic hedging
• Tie every conclusion to actual holdings (no hypothetical names)
• Respect the user's aggressive style — but call real risk plainly
• Tables for any 3+ data points
• Cite data freshness: "数据: clawock analyze-us / clawock analyze-hk {timestamp}"
• Flag stale legs loudly with ⚠️ before any conclusion drawn from them
• Do not substitute external "best practice" frameworks for the workspace data chain