Abel-Invest: Let Agents Run the Alpha Search

Where this fits

Your quant developer is asked: “Is there still alpha in momentum on intraday crypto?” Without a skill, the agent writes an ad-hoc backtest, overfits, and leaves no reproducible trail. abel-invest steps in as a skill that turns open-ended investing questions into a structured discovery workflow: resolve workspace state, create sessions and branches, run graph-enriched feature/model/ensemble search, enforce Abel Edge gates, and record evidence. From “can you improve this strategy” to “screen candidates for high Sharpe,” the agent has one reliable path.

Why agents benefit

• Activates on ambiguous phrasing like “a good strategy for X” or “is there alpha in Y,” because its description covers buy/sell, trading, backtesting, screening, and optimization even when the user never says “Abel.”
• Resolves workspace state first by reading evidence_ledger.json, frontier.md, agent_context.md, and exploration_path.md, so it avoids repeating explored branches and can continue prior work.
• Defaults to a reportable target—Sharpe above 2 and all Abel Edge gates passing—when no metric is specified, turning open-ended exploration into constrained search.
• Automates session and branch creation and records strategy rounds, producing an audit trail that makes debugging and human review concrete.
• Uses graph-enriched feature/model/ensemble search rather than hand-designed rules, so the agent explores a broader strategy space systematically.

Practical scenarios

“Is there alpha in volatility-scaled carry?”

The user asks a general alpha question. The agent activates abel-invest, checks the existing workspace, creates a branch, runs feature search on volatility-scaled carry, evaluates candidates through Abel Edge gates, and writes the outcome to the evidence ledger with Sharpe, return, and drawdown.

“Improve my current strategy’s Sharpe above 2”

The user returns with an existing workspace. The agent reads frontier.md and exploration_path.md, avoids failed branches, focuses modifications on the top frontier strategies, records the round, and updates the frontier for the next iteration.

“Screen 20 tickers for a report”

The user needs a reportable shortlist by morning. Using the skill’s default target, the agent screens candidates, keeps only those passing all gates, and produces an evidence ledger plus a ranked shortlist with metrics ready for human sign-off.

Add it to your agent workflow

Install with:

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

Then point your agent at it:

Use abel-invest to prepare a strategy discovery session for a long-only momentum strategy on S&P 500 constituents. Default target applies.

The agent resolves the workspace, creates a session, runs the search, and writes results to the evidence ledger.

Compare before adopting

Look at quality signals: does the generated evidence ledger contain enough detail to reproduce and debug a strategy? Does the skill gate strategies with statistical rigor or just print metrics? Check maintenance freshness: last commit, open issues, and maintainer responsiveness. Alternatives include general research skills that won’t enforce a discovery workflow, or your own bespoke scripts if you only need one-off studies. Also confirm workflow fit: abel-invest requires Python3 and expects you to follow its workspace state files.

Why it is worth tracking

abel-invest is worth watching because it is a focused, opinionated workflow, not just a collection of finance prompts. It translates broad requests into a reproducible search with explicit gates, which is rare in agent skills. With 19 stars it is still early, but the design pattern is useful and the project’s direction will tell you whether it becomes a standard quant-agent tool. Check its progress at https://www.openagentskill.com/skills/abel-ai-lab-abel-invest when evaluating your stack.