agentleFS
lucaswychanlucaswychan/quant-paper-agent/.agents/skills/quant-paper/SKILL.md

quant-paper

Skill0 starsChanged 3 months agoMITView on GitHub

Run or resume an autonomous, checksum-bound quantitative-finance paper workflow from discovery through visual analysis, audited replication or theory reconstruction, and explicitly requested research/paper-trading translation. Use as the primary front door when a user asks to find, download, analyze, replicate, reconstruct, implement, translate, continue, or resume an arXiv quant paper and wants the agent to complete all bounded local work with minimal intervention.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask only when the paper identity is materially ambiguous or a translation framework is missing.

    line 21
  2. Very specific:

    Start once, then follow the returned next_action until a terminal status:

    line 25
  3. Very specific:

    After completing the action, call qpa resume again.

    line 42
  4. Distinctive:

    Use the best scientifically meaningful classification: exact, otherwise approximate, otherwise illustrative.

    line 49
  5. Distinctive:

    Never route a theory reconstruction, failed replication, or inconclusive replication into translation.

    line 51
  6. Very specific:

    Treat complete, ineligible, and failed as terminal.

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Quant paper

Turn one bounded request into a resumable research workflow. Use qpa as the deterministic controller and the existing stage skills for semantic work. Do not ask for approval already conveyed by the request.

Resolve the goal

Infer the narrowest terminal goal from the user's verb:

  • find or search: discover;
  • download, save, or acquire: acquire;
  • analyze, extract, summarize, or explain: analyze;
  • replicate, reproduce, or backtest: replicate;
  • reconstruct a formula, theorem, boundary, or figure: reconstruct;
  • translate or implement for a named framework: translate.

Ask only when the paper identity is materially ambiguous or a translation framework is missing. Do not expand a narrower goal automatically.

Start or resume

Start once, then follow the returned next_action until a terminal status:

qpa run "<arXiv ID, URL, title, or query>" --goal <goal> --workspace "$QPA_WORKSPACE_ROOT" --json
qpa resume <workflow-id> --workspace "$QPA_WORKSPACE_ROOT" --json

Pass --paper-id after resolving a query, --search-results after saved discovery, --run-id when continuing an existing run, or --framework when the user supplies a missing framework. Never edit workflow.json by hand. Use --needs-input, --ineligible, or --failed with a concise code and --message only when the controller cannot derive the terminal outcome itself.

For each agent action, read the named stage skill completely before acting:

  • search_source or acquire_source: ../paper-search/SKILL.md;
  • extract_paper: ../paper-extract/SKILL.md and the available visual PDF capability;
  • execute_replication: ../paper-replicate/SKILL.md;
  • execute_reconstruction: ../paper-theory-reconstruct/SKILL.md;
  • translate_strategy: ../strategy-translate/SKILL.md.

After completing the action, call qpa resume again. Continue through recoverable validation, implementation, audit, comparison, and finalization work without returning control to the user.

Routing and defaults

  • Honor an exact paper, target, formula, framework, or exact-only constraint from the request.
  • For a generic replication request, use the controller-selected proposed target with the strongest primary-metric and net-of-cost evidence. If no metric target exists, reconstruct all visually verified core formulas.
  • For an explicit reconstruction request, use theory reconstruction even when metric targets also exist.
  • Use the best scientifically meaningful classification: exact, otherwise approximate, otherwise illustrative. If the user explicitly requires exact replication, stop instead of downgrading.
  • Translate only a finalized metric-bearing run whose verdict is replicated or partially replicated. Carry limitations into the translated research or paper-trading implementation.
  • Never route a theory reconstruction, failed replication, or inconclusive replication into translation.

Authorization boundary

A direct request authorizes the local stages needed for its goal, including planning, data acquisition from already allowed public sources, implementation, testing, comparison, and finalization. Bind each stage authorization to the immutable workflow request with --workflow-request <absolute-workflow.json> when recording approval or acceptance. Do not display hashes or ask for a second confirmation.

Pause only for unresolved paper identity, unavailable required data with no meaningful fallback, credentials, payment or licensing, a missing or unprobeable requested framework, publication, live trading, capital deployment, or scope expansion. Generated code is limited to research and paper trading.

Completion

Treat complete, ineligible, and failed as terminal. Report the final paper ID, achieved goal, selected target or formulas, run classification and verdict when applicable, artifact locations, and material limitations. For needs_input, ask only for the exact missing decision recorded by the controller.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.