openclaw-portfolio-thin

name: openclaw-portfolio-thin description: Orchestrates a thin OpenClaw portfolio workflow with fixed call sequence, fixed failure policy, and fixed output format. Use when user requests current market snapshot, DCA impact calculation, position updates in current_position.md, asset amount and total weight calculation, strategy opportunity scan from strategy/*.md, or explicit trade event phrases like "我今天卖了500股红利低波".

OpenClaw Portfolio Thin Skill

Scope

This skill is intentionally thin and only enforces:

1. Fixed call sequence
2. Fixed failure policy
3. Fixed response format

Do not invent new trading logic or override existing strategy parameters.

Source Of Truth

• Global function call contract: strategy/llm_function_call_contract.md
• Rebalance/agent execution hints: plan.md
• Snapshot generator: pull_snapshot.py
• Position file to read/update: current_position.md
• Strategy docs to scan: strategy/*.md

Priority order is fixed:

1. hard_risk_limits
2. strategy/llm_function_call_contract.md
3. strategy-specific preferences in strategy/*.md

Trigger Conditions

Use this skill when requests involve any of the following:

• Get latest market/portfolio status now
• Calculate DCA impact and update holdings
• Recompute per-asset amount and total allocation
• Scan strategy docs and report opportunities
• Natural-language trade events that imply position changes

Fixed Call Sequence

Always execute in this sequence. Do not skip or reorder.

Step -1: Bootstrap current_position.md (first use only)

Before any other step, check whether current_position.md exists in project root.

If file does not exist:

• Enter Q&A mode and ask user for:
  • cash (CNY, 万元)
  • holdings list (market, code/ticker, name, shares)
  • optional bucket mapping (if unknown, ask user to confirm UNCLASSIFIED)
• Generate current_position.md from user answers (do not invent holdings)
• Return status=data_needed if key fields are still missing after one clarification round
• Only continue to Step 0 after file creation succeeds

Step 0: Trade Event Detection (pre-check)

Before snapshot pull, detect whether the user message contains explicit trade events.

Examples:

• 我今天卖了500股红利低波
• 买入100股QQQ
• 减仓央企红利ETF 200股

If detected:

• Parse action, instrument, quantity, optional price/time
• Apply position delta to current_position.md first
• If required fields are missing, return data_needed with missing fields list

Step 1: Get Current Snapshot

• Read current_position.md, extract cash field first
• Generate/refresh portfolio_snapshot.json via existing workflow based on pull_snapshot.py
• Use contract function identity: get_portfolio_snapshot

Step 2: Compute DCA Impact And Position Update

• Compute DCA state from current_position.md + snapshot output
• Reflect DCA-caused position/cash delta in current_position.md when applicable
• Use contract function identity: compute_bucket_weights (for post-update aggregation)

Step 3: Compute Asset Amounts And Weights

• Compute each asset amount and total allocation from latest snapshot + positions
• Required outputs include per-asset CNY amount, total CNY value, and bucket weights

Step 4: Scan Strategy Docs For Opportunities

• Read strategy/llm_function_call_contract.md first, then scan other strategy/*.md
• Use market/position state to produce possible opportunities with evidence
• Contract-driven function chain (as needed): get_market_window -> get_cross_asset_features -> evaluate_gates -> propose_trade_actions -> validate_risk_checks

Step 5: Finalize Output

• Return strictly with the fixed JSON schema below
• Then append a short Chinese summary (3-6 lines) that only restates JSON conclusions

Fixed Failure Policy

Apply these rules exactly.

1. Cash parse failure:

   • Retry read/parse once
   • If still failed, use last successful cash value
   • Add error tag: cash_parse_failed

2. Snapshot pull failure:

   • Retry full pull once
   • If still failed, keep last valid snapshot and report failure in errors

3. US quote partial failure:

   • Preserve us_quote_errors
   • Do not abort global valuation if other assets are available

4. Missing required fields:

   • Return status=data_needed
   • Fill data_needed_if_any
   • Do not guess or synthesize missing values

5. Risk/constraint validation failure:

   • Allowed outputs only: hold or adjusted compliant actions
   • Never output executable non-compliant action

Fixed Output Format

Always output this JSON object first, then the short Chinese summary.

{
  "status": "ok|partial|data_needed|error",
  "as_of": "ISO-8601 datetime",
  "used_functions": [
    "get_portfolio_snapshot",
    "compute_bucket_weights",
    "get_market_window",
    "get_cross_asset_features",
    "evaluate_gates",
    "propose_trade_actions",
    "validate_risk_checks"
  ],
  "position_update": {
    "event_detected": true,
    "events_applied": [
      {
        "instrument": "512890",
        "action": "sell",
        "quantity": 500,
        "source_text": "我今天卖了500股红利低波"
      }
    ],
    "current_position_md_updated": true
  },
  "portfolio_metrics": {
    "per_asset_amount_cny": {
      "CN.512890": 0
    },
    "total_value_cny": 0,
    "bucket_weight": {
      "dividend": 0.0
    }
  },
  "opportunities": [
    {
      "strategy_id": "low_volatility_of_dividends",
      "instrument": "512890",
      "action_hint": "hold|buy|sell|reduce|increase",
      "confidence": 0.0,
      "evidence": [
        "来自 strategy 文档和当前仓位/行情的可审计证据"
      ]
    }
  ],
  "risk_validation": {
    "pass": true,
    "violations": [],
    "adjusted_actions": []
  },
  "errors": [
    {
      "code": "cash_parse_failed|snapshot_pull_failed|us_quote_partial_failed|...",
      "message": "string"
    }
  ],
  "data_needed_if_any": []
}

Rules:

• Keep used_functions in execution order.
• If a function is not run due to early data_needed, include only executed functions.
• Numeric fields must be numbers, not percent strings.
• confidence range: 0.0 to 1.0.

External Delivery (Optional)

If a file named .portfolio_webhook exists in the project root containing a webhook URL, pull_snapshot.py will automatically POST the generated markdown report to that URL after completion. This enables pushing reports to third-party chat bots (e.g., Feishu group bots). The file is gitignored to prevent accidental exposure.

Acceptance Checklist

• Normal path:

  • No manual trade event
  • Returns complete portfolio_metrics + opportunities
  • Sequence, failure handling, and output schema are stable

• Special keyword path:

  • Input includes explicit event (for example 我今天卖了500股红利低波)
  • Position update is applied first to current_position.md
  • Recalculation and opportunity output use updated positions