跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
金融

abel-invest

Use when the user asks how to invest, trade, buy or sell, find alpha, find or improve a trading strategy, backtest or stress a signal, screen candidates, optimize Sharpe/return/drawdown, run graph-enriched feature/model/ensemble search, or continue/prepare/debug an Abel strategy-discovery workspace — even if they don't say "Abel" and even when they just ask for "a good strategy for X" or "is there alpha in Y". When no metric target is specified, default to searching for a high-return, reportable strategy with Sharpe above 2 and all required Abel Edge gates passing. Prefer this over ad-hoc hand-designed strategy work.

Abel-ai-lab · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add Abel-ai-lab/predict-anything --skill abel-invest

技能说明


name: abel-invest description: > Use when the user asks how to invest, trade, buy or sell, find alpha, find or improve a trading strategy, backtest or stress a signal, screen candidates, optimize Sharpe/return/drawdown, run graph-enriched feature/model/ensemble search, or continue/prepare/debug an Abel strategy-discovery workspace — even if they don't say "Abel" and even when they just ask for "a good strategy for X" or "is there alpha in Y". When no metric target is specified, default to searching for a high-return, reportable strategy with Sharpe above 2 and all required Abel Edge gates passing. Prefer this over ad-hoc hand-designed strategy work. metadata: openclaw: requires: bins: - python3

Abel Invest Alpha Search

Use this skill for:

  • alpha search and candidate screening
  • continuing an existing Abel strategy discovery workspace
  • creating sessions and branches
  • preparing, debugging, recording, and reviewing strategy rounds
  • interpreting evidence_ledger.json, frontier.md, agent_context.md, and exploration_path.md

Activation Checklist

Always start by resolving workspace state before strategy work.

  1. Read references/workspace-bootstrap.md.
  2. Resolve the workspace location:
    • if alpha.workspace.yaml is in the current directory, use the current directory
    • else if abel-invest-workspace/alpha.workspace.yaml exists under the current directory, use that child workspace
    • else bootstrap a workspace before deep strategy work
  3. Run the active skill bootstrap shim for the resolved or default workspace: python3 <abel-invest-skill-root>/scripts/bootstrap_workspace.py --path <workspace-root>. Use --runtime-python /path/to/python only when the local machine cannot create a venv and the user intentionally provides an existing interpreter.
  4. Baseline-first: before from-scratch search, check whether a validated strategy for this target already exists in any baseline / strategy catalog the user maintains. If one exists, treat it as a benchmark and launchpad; iterate from it when useful rather than wasting rounds rediscovering it.
  5. If bootstrap reports auth_missing, use abel-auth, then rerun the active bootstrap shim. If it reports scaffold_stale, runtime_stale, env_missing, edge_missing, or edge_contract_missing, rerun the active bootstrap shim after fixing the stated blocker.
  6. Only start or continue session/branch work after bootstrap readiness is ready, unless the user explicitly asks you to inspect or repair setup.

Reference Routing

  • New workspace, workspace reuse, auth, generated-file refresh, or setup repair: read references/workspace-bootstrap.md.
  • New session, normal round loop, or resuming a session: read references/experiment-loop.md.
  • Live graph discovery, graph frontier expansion, or graph-informed alpha context: read references/discovery-protocol.md.
  • Selecting the default V3 graph or explicitly opting into a V4 release, or interpreting symbol versus canonical-node feeds: read references/graph-releases.md.
  • Creating or revising branch.yaml, reviewing evidence labels, path coverage, input realization, or exploration path use: read references/branch-authoring.md.
  • Writing engine.py, handling semantic/runtime failures, or checking temporal legality: read references/constraints.md.
  • Handling hosted paper contract requests, promoted strategy source edits, paper-contract-report.json, packaged strategy assets, or hosted paper state: read the emitted paper-contract-request.json first; read contractGuide.referencePath from this active skill when the request requires stateful continuation, source edits, or deeper gate diagnosis.
  • Explaining why the workflow is data-led, graph-informed, or evidence-boundary oriented: optionally read references/methodology.md.
  • Choosing concrete constructions while writing the engine: read references/proven-patterns.md (battle-tested patterns). Core path.
  • A hard Sharpe / MaxDD / PnL target is set: read references/guarded-optimization.md (performance-target search and reportability rules). Core path — not optional — when a performance bar is set.
  • Before writing "exhausted / ceiling / no edge": read references/experiment-loop.md and check the ledger requirements there.
  • Ordinary alpha search, data-driven candidate construction, or the next idea risks becoming another simple hand-written rule: read references/data-driven-construction.md before the first broad candidate. Core path.
  • No explicit metric target: use the normal experiment loop and default objective; do not treat this as a separate mode.

Operating Rules

Always:

  • Work workspace-first. Resolve workspace_root, research_root, and bootstrap readiness before session or branch work.
  • Reuse the default workspace when it already exists; reuse any resolved existing workspace before bootstrapping another one.
  • Bootstrap the workspace before deep strategy work when no workspace exists.
  • Use Abel Invest commands through the workspace command_prefix when available, not old aliases.
  • On a fresh install where abel-invest is not installed, run python3 <abel-invest-skill-root>/scripts/bootstrap_workspace.py --path abel-invest-workspace. Do not import abel_invest with the system interpreter for first-run bootstrap.
  • If a skill update changed the workspace scaffold, runtime contract, or generated workspace docs, rerun the active bootstrap shim before strategy work. Do not use workspace-local lifecycle commands to repair setup.
  • Reuse existing Abel auth first. If live access is missing, use abel-auth and rerun the active bootstrap shim.
  • Treat abel-auth/.env.skill as the normal shared auth/profile source. Workspace .env is only an explicit per-workspace override; do not copy API keys there unless the user intentionally wants this
阅读完整来源 ↗