Auto Paper Intake

Purpose

Create the run configuration and source inventory. Do not write manuscript prose or make claim decisions in intake.

Inputs

Collect or infer:

• paper_id
• target_venue
• target_name
• draft_path
• tex_roots
• bib_paths
• figure_paths
• figure_requirement_status
• materials_dir
• reference_paths
• experiment_evidence_index when docs/30_evidence/Experiment_Evidence_Index.{json,md} exists
• artifact_dir
• objective
• scope
• output_language
• compile_command
• human_gate_policy
• forbidden_directions

If a field cannot be inferred from files or user brief, write unknown and add a USER_GATE question to intake_report.md.

For blog, review, survey, or tutorial targets, set citation policy explicitly. Use citation_target_count: unknown or a positive target unless the operator explicitly asks for an uncited opinion memo.

Discovery

If the artifact directory is empty or missing, initialize the run scaffold first:

• .agents/skills/auto-paper/scripts/init_artifacts.py --paper-id <paper_id> --artifact-dir <artifact_dir> --workflow <workflow>

Templates are placeholders. Replace unknown values during intake and keep future-phase templates as scaffolds until their owning phase runs.

Use rg --files to find .tex, .bib, figures, notes, PDFs, Markdown files, CSV/JSON result files, and local reports. Index experiment evidence only through docs/30_evidence/Experiment_Evidence_Index.*; do not include iteration_log.json in source_index.md as a normal data source. If it is read for planning, mark it as a weak signal and cross-check with run artifacts or iteration reports. Prefer deterministic scripts when available:

• .agents/skills/auto-paper/scripts/init_artifacts.py
• .agents/skills/auto-paper/scripts/reference_inventory.py
• .agents/skills/auto-paper/scripts/tex_inventory.py
• .agents/skills/auto-paper/scripts/figure_requirement_scan.py
• .agents/skills/auto-paper/scripts/latex_guard.py

For PDF and Markdown materials, run or emulate figure_requirement_scan.py to search for figure/table cues. Record all matches in figure_requirement_scan.md with source locations. If the scan finds candidate figures or tables, set the next owner to $auto-paper-figure or layout and do not leave the need only in the raw PDF text.

If compile_command is present, run a non-mutating baseline compile or static guard. If absent, record compile_command: unknown.

Outputs

Write under auto_paper_output/<paper_id>/:

• config.yaml
• source_index.md
• tex_inventory.json
• intake_report.md
• experiment_source_map.md when an experiment evidence index is present
• figure_requirement_scan.md when PDFs, Markdown notes, or source materials were scanned for figure/table needs

Do not invent hidden controller state; write only the declared intake artifacts under auto_paper_output/<paper_id>/.

Gate Ledger

Report a Gate ledger entry with commands run, artifacts written, any USER_GATE or NOT_RUN reason, and the next owner before handoff.