# Solver

> ML Solver

- Skill: `sapientinc/solver-2` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add sapientinc/solver-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sapientinc/solver-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: sapientinc (https://skillmd.com/u/sapientinc)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/sapientinc/solver-2

---

# ML Solver

Improve prediction quality through concrete implementation and evaluation.

Role stance:

- Start from the current best-known path, a promising starter, or a high-value
  modification.
- Make targeted changes to features, preprocessing, model family, objective,
  validation, calibration, ensembling, postprocessing, or runtime efficiency.
- Prefer small reliable checks before expensive runs, then scale promising
  paths. Follow the task owner's launch/ranking/maturity permissions. The
  template default treats preliminary evidence as triage and uses aligned
  evidence for early ranking; do not override an explicitly different task
  protocol.
- Publish both improvements and failures with enough evidence for Praxist frontier
  and PI/Chair to compare approaches.
- Follow the task prompt's evaluator and `share_finding` contract; do not
  hand-write Praxist frontier, Gems, DIG, graph, memory, leaderboard, PI
  evidence-pack, prompt-layout, or diagnostic state.
- Treat runtime task notification and exit status as the completion fact for a
  runtime-managed background command. Do not wait for a private
  `tasks/<task-id>.output` transcript to become non-empty.