PubMed

Search and evaluate biomedical literature with effective queries, filters, and critical appraisal.

Security Scan

This is an instruction-only PubMed search and critical-appraisal guide; it requests no credentials, installs nothing, and its instructions are consistent with the stated purpose.

Purpose & Capability

The name and description ('Search and evaluate biomedical literature') match the SKILL.md content, which is a set of best-practice rules for constructing PubMed queries and appraising studies. No unrelated binaries, env vars, or installs are requested.

Instruction Scope

SKILL.md provides high-level, domain-specific guidance (MeSH terms, filters, appraisal heuristics). It does not instruct the agent to read local files, exfiltrate data, call arbitrary external endpoints, or access credentials. References to external resources (Retraction Watch, Unpaywall) are advisory only.

Install Mechanism

There is no install specification and no code files — the skill is instruction-only, which minimizes on-disk risk.

Credentials

The skill requests no environment variables, credentials, or config paths. That is proportionate for an advice-only PubMed helper.

Persistence & Privilege

always:false and default autonomous invocation are unchanged. The skill does not request persistent presence or elevated privileges and does not modify other skills or system settings.

Assessment

This skill is a text-only guide for searching and appraising PubMed literature and appears internally consistent and low-risk. Note the registry/source fields are empty (no homepage, unknown source); if provenance matters to you, prefer an official or well-known source. Because it contains no code, installing it does not add executables or request secrets — still avoid pasting private credentials into prompts and be cautious if a future version adds network calls or install steps (review any added install scripts, URLs, or required env vars before approving).

PubMed Research Rules

Query Construction

• Use MeSH terms for precise searching — controlled vocabulary ensures you find related concepts regardless of wording
• Boolean operators must be uppercase: AND, OR, NOT — lowercase is ignored
• Phrase searching with quotes: "heart failure" not heart failure — unquoted searches terms separately
• Field tags narrow searches: [Title], [Author], [MeSH Terms] — example: aspirin[Title] AND prevention[MeSH]
• Truncation with asterisk: therap* finds therapy, therapies, therapeutic

Essential Filters

• Article type matters: Clinical Trial, Systematic Review, Meta-Analysis — filter by study design
• Publication date for recent evidence — older studies may be superseded
• Free full text filter if access is limited — but don't ignore paywalled high-quality studies
• Humans filter excludes animal studies — relevant for clinical questions
• Language filter if translation isn't feasible

Study Hierarchy

• Systematic reviews and meta-analyses synthesize multiple studies — start here for established topics
• Randomized controlled trials (RCTs) are gold standard for interventions — but not all questions are answerable by RCT
• Cohort studies for long-term outcomes and rare exposures
• Case-control for rare diseases
• Case reports are lowest evidence — interesting but not generalizable
• Guidelines synthesize evidence into recommendations — check who wrote them and when

Critical Appraisal

• Sample size matters — small studies may show effects that don't replicate
• Check confidence intervals, not just p-values — narrow CI with meaningful effect size beats p<0.05
• Funding source and conflicts of interest affect interpretation — industry-funded studies favor sponsors
• Primary vs secondary outcomes — cherry-picking significant secondary outcomes is common
• Intention-to-treat vs per-protocol analysis — ITT is more conservative and realistic

Common Traps

• Abstract conclusions may oversell results — read methods and results sections
• Single studies rarely settle questions — look for replication and systematic reviews
• Statistical significance isn't clinical significance — 1% improvement may not matter to patients
• Retracted papers still appear in searches — check Retraction Watch for controversial papers
• Predatory journals publish low-quality research — verify journal reputation
• Preprints haven't been peer-reviewed — useful for speed but not vetted

Search Strategy

• PICO framework: Patient/Population, Intervention, Comparison, Outcome — structures clinical questions
• Start broad, then narrow with filters — missing relevant papers worse than sorting through extras
• Save searches for ongoing monitoring — PubMed can email when new papers match
• Related Articles feature finds similar papers — useful after finding one good paper
• Citation tracking: who cited this paper? — follow research forward in time

Evaluating Sources

• Impact factor indicates journal prestige, not individual paper quality
• First and last authors typically did the work and led the project
• Corresponding author handles questions — contact for clarifications
• Check author affiliations — institutional reputation matters
• Methods section determines if results are trustworthy — results are only as good as methods

For Specific Questions

• Treatment efficacy: RCTs and systematic reviews first
• Diagnosis accuracy: sensitivity/specificity studies
• Prognosis: cohort studies with long follow-up
• Etiology/harm: cohort or case-control studies
• Prevention: RCTs when available, cohort otherwise

Practical Tips

• PubMed is free but full text often isn't — check institutional access, Unpaywall, or request from authors
• PMID is the unique identifier — use it for precise citations
• Export to reference manager (EndNote, Zotero) — manual citation is error-prone
• Clinical Queries filter pre-filters for clinical relevance — therapy, diagnosis, prognosis, etiology
• Similar Articles and Cited By expand discovery — algorithm finds related work

Red Flags in Papers

• No control group for intervention studies
• Conclusions not supported by data presented
• Missing or inadequate statistical analysis
• Selective reporting of outcomes
• Conflicts of interest not disclosed
• Extraordinary claims without extraordinary evidence