Canonicalize literature sources into kb/library/literature/, including PDFs from raw/, arXiv or OpenReview links, paper pages, blogs, and project pages. Use when Codex needs to download or stage source material, deduplicate by hashes or metadata, move buffered PDFs into the library, resolve duplicate reviews, or refresh the shared literature graph and metadata files before note authoring.
Treat raw/ as immutable intake evidence and library/literature/ as the canonical llm-wiki store.
ingest (owner): stage local PDFs or URLs, deduplicate, canonicalize, and materialize library/literature/<source-id>/.query (limited): run duplicate/fingerprint lookups and extraction sanity checks needed for ingest decisions only.lint (limited): validate ingest-time schema completeness for metadata.yaml, claims.yaml, and methods.yaml.index (owner, ingest-time only): refresh library/literature/index.yaml, graph.yaml, and ingest-time keyword registration after successful ingest.log (owner): write intake provenance and duplicate-review records so later skills can audit decisions.research-note-author), or long-term taxonomy cleanup (literature-tagger).kb/intake/papers/downloads/<intake-id>/.raw/ or intake staging in place; only copy/move with hash checks and provenance.intake/papers/review/pending.yaml and stop canonical creation.library/literature/<source-id>/source/, then refresh index and graph.research-note-author.literature-tagger instead of stretching ingest logic.metadata.yaml, claims.yaml, methods.yaml, and index entries.inputs with staged source paths or source URLs whenever possible.research-conductor.canonical_url and site_fingerprint instead of leaving local-buffer placeholders.short_summary into canonical metadata so the library index stays searchable.claims.yaml as a placeholder scaffold by default: mark it unverified and avoid presenting it as manually validated extraction.topics and tags from kb/memory/domain-profile.yaml, then register discovered keyword tags for downstream retrieval.literature-tagger.intake/.../pending.yaml, metadata.yaml, and index.yaml), not left in chat-only text.PyYAML or PDF backend is missing, fail early and hand back to research-conductor for runtime recovery.python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --source raw/example.pdf
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --source raw/example.pdf --program-id my-program
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --source "https://arxiv.org/abs/2501.09747"
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --search-result kb/library/search/results/latest-vla.yaml
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --search-result kb/library/search/results/latest-vla.yaml --program-id my-program
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py resolve-review --review-id paper-review-... --decision existing --canonical-id lit-...
python3 .agents/skills/literature-corpus-builder/scripts/ingest_literature.py refresh-claims --source-id lit-arxiv-2501-09747v1
python3 .agents/skills/research-conductor/scripts/run_with_runtime.py .agents/skills/literature-corpus-builder/scripts/ingest_literature.py ingest --source raw/example.pdf
cat kb/programs/my-program/workflow/reporting-events.yaml
python3 .agents/skills/research-note-author/scripts/prepare_note_assets.py prepare-literature-note --source-id lit-arxiv-2501-09747v1
raw/ or intake staging directories.claims.yaml as verified evidence.note.md; that belongs to research-note-author.literature-tagger.intake/papers/review/pending.yaml before declaring ingest complete.raw/ imports were canonicalized with provenance and no orphaned buffer artifacts remain.canonical_url, site_fingerprint, short_summary, and inputs on new entries.library/literature/index.yaml and graph.yaml.research-note-author and note.md completion.skill-evolution-advisor.