This source did not publish a separate summary. Review SKILL.md before using the skill.
SKILL.md
five-questions
Systematically analyze an empirical economics paper using the five-question framework.
Produces a structured Chinese report (2000–3000 chars) with a transferable methodology checklist.
Input
/five-questions path/to/paper.pdf
Also accepts: /five-q path/to/paper.pdf or natural language trigger.
File locations (explicit)
This skill keeps temporary intermediate files inside the current project, not next to the source PDF, so the user's working tree stays tidy and the workspace is easy to find or clean up.
Disposable; safe to delete after the final report is generated
Final report
${CLAUDE_PROJECT_DIR}/{paper_slug}_five_q.md
Persistent deliverable
paper_slug = lowercase filename of the source PDF with spaces replaced by underscores and the .pdf extension stripped. timestamp = YYYYMMDD_HHMMSS.
If ${CLAUDE_PROJECT_DIR} is unset (rare — happens when the skill is invoked outside a project context), fall back to the current working directory (os.getcwd()).
Orchestration
Follow these steps exactly:
Step 1 — Validate Input
Parse the PDF path from args. If not provided, ask the user for the path.
import os
pdf_path = args.strip()
if not os.path.exists(pdf_path):
print(f"ERROR: PDF not found: {pdf_path}")
sys.exit(1)
Step 2 — Create Work Directory
Place the temporary workspace inside the current project directory (NOT next to the PDF):
Run the conversion script bundled with this skill, using the sci conda environment:
conda run -n sci python3 \
"${CLAUDE_PROJECT_DIR}/.claude/skills/five-questions/scripts/convert_pdf.py" \
"{pdf_path}" "{work_dir}/paper.md"
If markitdown is not installed: conda run -n sci pip install 'markitdown[pdf]'
Check that {work_dir}/paper.md exists and has >2000 characters.
If <2000 chars, warn the user ("可能是扫描版PDF，文本提取有限") but continue.
Step 4 — Dispatch 5 Parallel Q-Agents
Send ALL five Agent tool calls in a SINGLE message (mandatory for parallel execution).
Each agent reads its role definition document first, then reads the paper and writes its answer.
Prompt template for each agent (substitute N, name, output path per agent):
