القائمة
Run experiments based on the plan from read-exp. Reads check_exp.json and check_code.json, creates environment, runs experiments, compares against paper numbers.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Organize a paper PDF and its codebase into clean reading artifacts. Use this before any other read-* skills.
Verify each cited reference exists via web search.
Deep extraction of all verifiable claims from a research paper. Actively proposes implicit claims, splits compound statements, and flags cross-reference inconsistencies.
Verify paper claims with evidence_type "code" against the actual codebase. Use after read-txt to check if code matches what the paper says.
Review experiment design and produce a prioritized reproduction plan. Targets claims with evidence_type "experiment" from check_claim.json. Output feeds into run-exp.
Web-search validation of citation and novelty claims from check_claim.json. Checks whether cited facts are accurate and novelty claims ("first", "pioneering") hold against published literature.
| name | run-exp |
| description | Run experiments based on the plan from read-exp. Reads check_exp.json and check_code.json, creates environment, runs experiments, compares against paper numbers. |
| author | PaperDoctor Research |
| license | MIT |
| argument-hint | ["paper_dir"] |
| dependencies | ["read-exp","read-code"] |
| allowed-tools | Read, Grep, Glob, Bash |
This skill requires explicit user approval before execution. Do not start automatically after Phase 2 — present the reproduction plan summary and wait for the user to confirm.
{paper_dir}/reports/check_exp.json — experiment plan from read-exp (priority, feasibility, commands){paper_dir}/reports/check_code.json — code issues from read-code (fix before running)- [ ] Step 1: Create environment
- [ ] Step 2: Run experiments
conda create -n paperdr-{paper_stem} python=<version> -y
conda activate paperdr-{paper_stem}
# follow README for install
Download data/checkpoints as needed. Fix any warning/error items from check_code.json before running.
Run experiments from check_exp.json in priority order (high first). Prefer eval over training.
For each experiment, compare output against the paper's numbers. Update check_exp.json with:
{
"our_result": {"metric": 0.0},
"status": "pass",
"note": "..."
}
| Status | Meaning |
|---|---|
pass | Matches paper within ~1-2% |
warning | Partial match or needed fixes |
error | Significant difference or could not run |
error — note cause, do not retry endlessly