The first version looked for node_modules in the current release, which does not work when the same
commit is deployed again (its directory is replaced). Cache hard-linked installs under
/opt/llm-survey/cache/node_modules/<lock hash> instead.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FbMbJPadPzVwvt9rU9GBNo
The data build (download, ingest, work types: about 5 of the 8 minutes), npm ci, the Next.js compiler
cache and the Postgres load are now cached under /opt/llm-survey/cache and redone only when their
inputs change. Raw data is refreshed by a nightly run or by a manual run with refresh_data.
Docs-only pushes and pushes already superseded by a newer commit on main skip the deploy.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FbMbJPadPzVwvt9rU9GBNo
- /api/articles[/slug]: GET/PUT/POST/DELETE, article JSON stored in PostgreSQL (schema analysis)
- /api/tool-taxonomy: categories (tree), per-paper tools, each tool in one or more categories
- /contents/tools-of-agents: category table with roll-up and links to the papers; removed
"By kind of contribution" and "Recent papers"
- content moves out of the repo: removed content/articles, load-articles.js, agent-tools.ts
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FbMbJPadPzVwvt9rU9GBNo
- Work type = area x role (fixed vs varied), vocabulary in scripts/worktypes.json
- LLM-read labels for 2,678 agent-related papers (data/labels), rule labels for the rest
- Per-type synthesis notes (why-same, sub-variants, benchmarks, workflow)
- Pages: /work-types matrix and detail (timeline, benchmarks, workflow, score charts), /benchmarks
- build_worktypes.py adds work-type tables to data/papers.db
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bhs9LU8eMixk6g1kb1d3j7
- Next.js/React/TypeScript app: overview, paper explorer with FTS5 search,
paper detail, global taxonomy tree with per-conference tag breakdown
- Python ingestion (ICLR/ICML 2026 via paperlists, ACL Anthology 2026) into
data/papers.db, plus keyword-scored taxonomy (scripts/taxonomy.py)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bhs9LU8eMixk6g1kb1d3j7