Skip to content

marciob/skill-research-papers / research-papers

research-papers

Verified

Research, filter, compare, and synthesize scientific papers across arXiv, OpenAlex, ACM Digital Library, DBLP, Semantic Scholar, PubMed, SSRN, and publisher landing pages. Use when the agent needs to do literature discovery, find top or recent papers, build a reading list, compare methods or results, trace citation graphs, summarize evidence, or produce properly attributed paper recommendations and literature reviews.

FileKindSize
SKILL.mdskill md14.5 KB
agents/openai.yamlconfig221 B
references/api-playbook.mdmarkdown16.9 KB
references/output-shapes.mdmarkdown2.5 KB
references/screening-rubric.mdmarkdown3.0 KB
references/smoke-tests.mdmarkdown4.4 KB
references/source-guides.mdmarkdown4.9 KB
scripts/.gitignoreother7 B
scripts/fetch_and_parseexecutablescript990 B
scripts/fetch_and_parse.pyexecutablescript52.0 KB
scripts/requirements.txtmanifest496 B
scripts/smoke-test-sources.shexecutablescript5.8 KB

references/screening-rubric.md

View on GitHub
# Screening Rubric

Use this file when ranking papers, selecting a shortlist, or building comparison notes.

## Ranking Dimensions

Score papers qualitatively, not mechanically. A simple high-medium-low judgment is usually enough.

### Direct relevance

High:

- The paper directly answers the user's topic, method, or domain question

Low:

- The paper is adjacent but not central

### Evidence quality

High:

- Full paper read
- Clear methods and evaluation
- Claims supported by data or careful argument

Low:

- Abstract-only access
- Vague methods
- Strong claims with weak support

When full text is available, always read it before finalizing the ranking. Abstract-only assessments should be treated as provisional until confirmed by full-text reading.

### Methodological value

High:

- Introduces a strong method, benchmark, dataset, or conceptual framing

Low:

- Incremental change with limited explanatory value

### Influence

High:

- Widely cited, well known, or a common anchor in the field

Low:

- Hard to place in the literature and not obviously impactful

### Recency

High:

- Materially reflects the current state of the field

Low:

- Old enough that later work likely changed the picture, unless it is seminal

### Publication status

High confidence:

- Peer-reviewed or clearly published in a reputable venue

Lower confidence:

- Preprint or working paper with no confirmed publication

## Selection Rules

- Keep a mix of seminal and recent papers when the user asks for "top" papers.
- Include a survey or review paper when it helps orient the user quickly.
- Do not let citation count dominate direct relevance.
- Separate preprints from published work in the final presentation.
- Prefer papers with accessible full text when the task requires detailed synthesis.
- Note which papers have resolved full-text URLs. Prioritize reading those during evidence extraction.
- When two papers are otherwise equal in relevance, prefer the one with full-text access.

## Note Template

Use a compact note for each shortlisted paper:

- Citation:
- Type: published paper, preprint, review, benchmark, dataset, working paper
- Problem:
- Approach:
- Data or benchmark:
- Main finding:
- Limitation:
- Why included:
- Confidence: full text, abstract only, or metadata only

## Comparison Rules

- Do not compare headline numbers across different datasets or evaluation setups without stating the mismatch.
- Do not equate publisher prestige with result quality.
- Distinguish "important historically" from "best current evidence."
- Mark when a recommendation is based on citation graph signals rather than deep reading.

## Search Expansion Rules

If the candidate set is too thin:

- Follow references and citations from two or three anchor papers
- Add synonyms and older terminology
- Add adjacent benchmark or dataset names

If the candidate set is too broad:

- Narrow by task, population, benchmark, or date
- Separate surveys from primary studies
- Restrict to one domain or publication type
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexUnverifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall