mathflow

Installed Root

Resolve the installed reasflow-dev skills root before running packaged scripts:

REASFLOW_SKILLS_ROOT="${REASFLOW_SKILLS_ROOT:-}"
if [ -z "$REASFLOW_SKILLS_ROOT" ]; then
  if [ -d ./.agents/skills ]; then
    REASFLOW_SKILLS_ROOT="$(pwd)/.agents/skills"
  elif [ -d "$HOME/.agents/skills" ]; then
    REASFLOW_SKILLS_ROOT="$HOME/.agents/skills"
  else
    echo "reasflow-dev skills not found in ./.agents/skills or $HOME/.agents/skills" >&2
    exit 1
  fi
fi

Role

mathflow is the only top-level skill for staged mathematical work. It is a router and state tracker, not a request to eagerly read every stage guide.

Use the skill tool only for mathflow; do not load stage files as top-level skills.

Use when

Use this skill for mathematical research, contest modeling, derivations, proof checks, numerical experiments, result validation, or final mathematical reporting when the work may need staged judgment.

Stage Files

Stage guidance lives in subordinate files under this skill directory:

• stages/problem-analysis.md
• stages/mathematical-modeling.md
• stages/derivation-and-proof-checking.md
• stages/research-planning.md
• stages/numerical-experimentation.md
• stages/result-validation.md
• stages/self-audit-loop.md
• stages/report-writing.md

When the next step requires detailed stage rules, read only the one relevant stage file. Do not read all stage files up front.

Mathflow State

Start each staged response with a compact state block:

Mathflow State
- Current stage: <stage name or "triage">
- Evidence so far: <one short sentence>
- Next stage: <one stage name or "fast path">
- Why this stage: <one short sentence>
- Needs from user: <missing input or "none">

For small tasks, use fast path and answer directly after the state block when a full stage handoff would add ceremony without reducing risk.

Routing Rules

• Start underspecified or open-ended work in problem-analysis.
• Use mathematical-modeling before introducing new variables, equations, assumptions, or model families.
• Use derivation-and-proof-checking when a claim depends on proof, derivation, symbolic manipulation, or mathematical argument quality.
• Use research-planning before simulations, solver runs, sweeps, empirical checks, or numerical comparisons.
• Use numerical-experimentation for reproducible computational execution after an objective, baseline, metric, and stop rule exist.
• Use result-validation when a result needs special cases, limit cases, sensitivity, consistency, or counterexample checks.
• Use self-audit-loop only as the final adversarial review before strong conclusions or final reporting.
• Use report-writing only when prior work has enough grounded material to write honestly.