Skip to content

sidiangongyuan/codex-skills-library / skills/research-evidence

research-evidence

Verified

Use when research writing, reviews, rebuttals, or related-work planning needs academic evidence. Searches literature, screens candidate papers, audits novelty and missing work, verifies DOI/BibTeX/reference metadata and citation authenticity, and checks whether sources support the stated claims.

FileKindSize
LICENSEother1.0 KB
SKILL.mdskill md5.0 KB
agents/openai.yamlconfig277 B
references/recent-literature-audit.mdmarkdown3.3 KB
references/tooling.mdmarkdown2.9 KB
scripts/arxiv_query_matrix.pyscript8.3 KB

references/tooling.md

View on GitHub
# Research Evidence Tooling

## Dedicated Environment

Use only the dedicated environment:

```bash
${CODEX_HOME:-$HOME/.codex}/tools/research-evidence/bin/paper-search --help
${CODEX_HOME:-$HOME/.codex}/tools/research-evidence/bin/academic-refchecker --help
```

Pinned packages:

- `paper-search-mcp==0.1.3`
- `academic-refchecker==3.0.147`

Do not use global `uv tool install` or assume these commands are on the global `PATH`.

## Paper Search

`paper-search-mcp==0.1.3` installs importable searchers but no console scripts, so this environment includes a local `paper-search` wrapper in the venv `bin/` directory.

Examples:

```bash
${CODEX_HOME:-$HOME/.codex}/tools/research-evidence/bin/paper-search \
  "collaborative perception CVPR infrastructure vehicle" \
  --source arxiv \
  --max-results 5

${CODEX_HOME:-$HOME/.codex}/tools/research-evidence/bin/paper-search \
  "multimodal autonomous driving ICLR" \
  --source arxiv \
  --max-results 10 \
  --format json
```

Available wrapper sources: `arxiv`, `pubmed`, `biorxiv`, `medrxiv`.

Use arXiv first for CV/ML metadata smoke checks. For high-stakes
recent-work or novelty-risk audits, do not rely on `paper-search` as the only
search route: wrapper results can be noisy, can bury exact recent competitors,
and may miss terminology variants. Prefer a direct arXiv API query matrix and
use `paper-search` as a smoke or secondary cross-check.

The skill also includes a standard-library helper:

```bash
python "<skill-dir>/scripts/arxiv_query_matrix.py" \
  --domain-term "your domain term" \
  --method-term "your method term" \
  --task-term "your task term" \
  --neighbor-term "alternate terminology" \
  --since-year 2024 \
  --max-results 5 \
  --dry-run
```

Resolve `<skill-dir>` to the installed `research-evidence` directory before
running the command; do not assume a particular global skills root.

Pass `--run` only when live arXiv metadata is needed. The helper prints to
stdout and does not write report files by default.

Do not use Google Scholar scraping by default. Do not download PDFs unless the
user explicitly asks and the source is open access.

The environment also includes `paper-search-mcp`, a local wrapper for starting the package's stdio MCP server, but Codex MCP is not registered in v1.

## RefChecker

Default non-LLM invocation:

```bash
${CODEX_HOME:-$HOME/.codex}/tools/research-evidence/bin/academic-refchecker --paper /path/to/paper-or-bib
```

Allowed input types include local PDF, local LaTeX, local text references, BibTeX, arXiv ID, or URL. Prefer user-provided local files for submission checks.

Do not pass these flags unless the user explicitly requests LLM-assisted checks:

- `--llm-provider`
- `--hallucination-provider`
- `--llm-model`
- `--hallucination-model`

Do not pass `--output-file` or `--report-file` unless the user explicitly requests saved artifacts.

Use `--semantic-scholar-api-key` only if the user has provided or configured a key.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall