ADEQC Paper Analysis

Audit research papers with evidence chains — not vibes.

A skill for Cursor, Claude Code, and OpenAI Codex. Analyze papers with ADEQC_Law (Advanced Decision Evaluation Logic Chains): classify every finding as explicit (Ze) / implicit (Zi) / hype (I) evidence, score hypotheses W(h|D,B), and assess true novelty and experimental rigor against the paper's own code repo.

Install

npx adeqc-paper-analysis install
npx adeqc-paper-analysis install -g
npx adeqc-paper-analysis install cursor claude-code

| Scope | Flag | Paths | |-------|------|-------| | This repo | (default) | .cursor/skills/, .claude/skills/, .agents/skills/ | | All your projects | -g | ~/.cursor/skills/, ~/.claude/skills/, ~/.agents/skills/ |

Run without agent names to pick from an interactive list (↑↓ move, Space toggle, Enter confirm). install -g with no names updates agents already installed globally; if none exist yet, you pick which to add.

Restart your agent after installing.

Uninstall

npx adeqc-paper-analysis uninstall
npx adeqc-paper-analysis uninstall -g
npx adeqc-paper-analysis uninstall cursor

Uninstall with no agent names removes the skill from all agents in that scope.

Does not delete Analysis/, repos/, or .modus-querens/ artifacts next to your papers.

Usage

1. Drop a paper URL, arXiv id, title, or local PDF on your agent.
2. Ask something like "ADEQC this paper", "analyze the novelty of …", or "run Paper_Breaker on …".
3. The agent resolves → downloads the paper → clones the official repo → runs the evidence-chain analysis → writes Analysis/{paper_id}/{paper_id}_Pure.md.

Phase 2 — corpus cross-reference

Once ≥10 papers are analyzed, cross-proof them against each other (external supports, counter-evidence, prior-art inversion). Requires the modus-querens skill for tree indexing + BM25 recall:

npx modus-querens install -g
uv run --with rank-bm25 python <skill>/scripts/prepare_crossref_run.py \
  --analysis Analysis/ --run .modus-querens/runs/<date>-adeqc-crossref

What's in the skill

| File | Role | |------|------| | SKILL.md | Workflow: resolve → acquire → repo recon → 3-agent analysis → scoring → deliverable | | reference.md | W formula, novelty/rigor rubrics, classification edge cases | | template.md | _Pure.md deliverable skeleton | | examples.md | Condensed worked example | | crossref.md | Phase 2 corpus proofing workflow | | scripts/ | Probe extraction + cross-ref run prep (BM25 via modus-querens) |

License

MIT — see LICENSE.