linxichen/dsh-rigorquant
RigorQuant preset + skill pack: unattended walled multi-agent research for empirical and computational mathematics (economics, finance, portfolio), with a four-part pre-implementation check battery and a jacobian/Lean escalation lane.
RigorQuant is an agent preset + bundled skills that turns one DeepSeek Harness session into a context-isolated multi-agent research lab for empirical/computational mathematics — economics, finance, portfolio construction/optimization and simulation. Parallel blank-context explorers propose candidate methods; an OffGridThinker works fully isolated (raw model + compute tools, no web/skills/other results); a ground-truth track re-derives analytic closed forms twice by different means; an adversary eliminates routes by counterexample only; a four-part check battery (closed-form equality, exact invariants, analytic bounds, statistical hardening) runs BEFORE numerical implementation; and a meta-validator (rq_check.py) refuses a PASS whose evidence is missing — reading the audit record, never the study's own claims. Fixed-seed + LLN conventions govern stochastic work; PASS auto-implements and proceeds, BLOCKED runs 3 rounds on the same gap, BUDGET checkpoints after 5 rounds. The framework runs unattended within one live session only — crossing a session boundary disarms the goal.
Install
dsh plugin --profile web add dsh-rigorquantnpm dsh-rigorquant 0.4.1 verified 2026-09-03 (npm owner linxichen == repo owner; repository field empty so ownership is owner-name-consistent). README install is three routes — one-line npx dsh-rigorquant (preset + compute lane + plugin), clone + ./install.sh, or the ecosystem bundle path dsh plugin --profile web add dsh-rigorquant (requires dsh >= 0.1.2-alpha.1; the bundle's rq-preset-sync row lands the agent preset + compute lane on next profile start).
Compatibility
DSH >= 0.1.2-alpha.1 (native child agentOptions.reasoningEffort); compute lane uses sympy, numpy, mpmath, cvxpy, hypothesis, jax; optional Lean checkers; opt-in jacobian MCP escalation lane.
Details
- Repo: linxichen/dsh-rigorquant
- Category: Skills
- Stars: 0
- Version: npm dsh-rigorquant 0.4.1 (agent preset + skills + plugin bundle)
- Last push: 2026-09-10
- First seen: 2026-08-14
Recent updates
0.4.1 current on npm (verified 2026-09-03).
FAQ
- What kind of research is it for?
- Empirical/computational mathematics with verifiable claims — economics, finance, portfolio construction/optimization, simulation, computational econ/finance — where results can be checked by re-derivation and counterexample.
- Who checks the research?
- An eight-role separation enforced by composition: the producer never checks its own work. Explorers propose, an adversary eliminates by counterexample, a double-checker re-derives load-bearing claims twice, and rq_check.py refuses PASSes with missing evidence.
- Does it run unattended forever?
- No — it runs unattended within one live session; crossing a session boundary disarms the goal and one human 'continue' turn re-arms it.
Alternatives
guard42/dsh-humanize · jeremy9682/dsh-skill-pack