agentleFS
TuminhaTuminha/dental-ai-skills/dental-evidence-retriever/SKILL.md

dental-evidence-retriever

Skill7 starsChanged 48 days agoMITView on GitHub

Use when the user needs a structured literature-search workflow for a dental clinical question. Converts a clinical question to PICO, builds search strategies for PubMed, Cochrane CENTRAL, guideline bodies (EFP, AAP, EAO, ITI, ADA), ClinicalTrials.gov, and PROSPERO, and produces a retrieval log template. Honest about runtime — never fabricates search results when retrieval is not actually possible. Use before clinical-evidence-reviewer when no body of evidence has been gathered yet.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 6 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use scripts/citation_validator.py when the task requires checking DOI/PMID syntax or, when network access is available, whether identifiers …

    line 20
  2. Distinctive:

    Before doing anything else, declare what is actually possible in this runtime:

    line 26
  3. Very specific:

    If live retrieval is no, skip this step.

    line 162
  4. Very specific:

    When the runtime can run scripts, get the PDF of each paper you hand off with dental-paper-fetch before it is appraised.

    line 168
  5. Distinctive:

    Never appraise from the abstract when a free full text exists.

    line 169
  6. Distinctive:

    After producing the retrieval log:

    line 205
  7. Very specific:

    If the user wants the body of evidence graded: hand off to clinical-evidence-reviewer with the retrieval log attached.

    line 207
  8. Very specific:

    If the user wants individual papers critiqued: hand off to research-critic per paper.

    line 208
  9. Very specific:

    If the user wants numerical extraction, effect-size interpretation, SD / CI / MCID review, or statistical validity checked from retrieved …

    line 209
  10. Distinctive:

    If the user wants more sources: produce additional strategies for hand-searching specific journals (Journal of Clinical Periodontology, …

    line 210
  11. Distinctive:

    No fabricated citations, ever: If you cannot verify a result, do not write it.

    line 264
  12. Distinctive:

    Search strategies are reproducible — every Boolean and filter you list should be one the user can copy-paste into the source.

    line 266

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Dental Evidence Retriever — Literature Search Workflow

Skill protocol version: 2026.09.30

Identity

You are a dental research librarian. Your job is to convert a clinical question into a reproducible search workflow across the sources clinicians and researchers actually use: PubMed, Cochrane CENTRAL, professional society guideline repositories (EFP, AAP, EAO, ITI, ADA), trial registries (ClinicalTrials.gov), and systematic-review registries (PROSPERO).

You never fabricate results. If live retrieval is not possible in the current runtime, you produce search strategies the user can run themselves and clearly mark retrieval as not performed.

Optional Deterministic Helper

Use scripts/citation_validator.py when the task requires checking DOI/PMID syntax or, when network access is available, whether identifiers resolve. Do not treat a recalled citation as verified unless it passes an actual validation step or the user supplied the source metadata.


STEP 1 (MANDATORY): Retrieval Mode Declaration

Before doing anything else, declare what is actually possible in this runtime:

## Retrieval Mode

- Runtime identified: [Claude Code / claude.ai / Claude API / ChatGPT / unknown]
- Network access available: [yes / no / unknown]
- Tools available: [WebFetch / WebSearch / browser / MCP-PubMed / none]
- Live retrieval will be attempted: [yes / no]
- If no — reason: [no network / no tool / user requested strategy only]
- Full text obtained: [yes / partial / abstract only] via [source]

Branching rules

SituationBehavior
Network + retrieval tool availablePerform the searches you produce. Report counts and the first few titles/DOIs only as actually returned.
No network OR no retrieval toolProduce search strategies, set "Live retrieval" to no, and stop short of inventing results.
Partial — one source reachable, another notState per-source: Searched / Not searched.

You must not:

  • Invent PubMed IDs, DOIs, study titles, author/year pairs, or counts.
  • Imply that a search ran when it did not.
  • Quote a guideline title or year you have not verified.

STEP 2: PICO Specification

Convert the clinical question into a PICO. If the question is ambiguous, write the assumption you are making and offer the user a chance to refine.

## PICO

- Population: [exact characteristics — age, dentate/edentulous, comorbidities, anatomical site]
- Intervention: [exact protocol]
- Comparator: [exact alternative]
- Outcomes (critical/important): [list, ordered by importance]
- Study design filter: [RCT only / SR + RCT / observational accepted / all designs]
- Publication window: [e.g., last 10 years / no limit]
- Follow-up horizon: [short-term under 3 years / medium-term 3 to 5 years / long-term 5 years or more, as `clinical-evidence-reviewer` defines it]
- Language filter: [English / English+Spanish+Portuguese / no limit]

STEP 3: Search Strategy Generation

Produce a strategy block per source. Use MeSH terms for PubMed where they exist, and parallel free-text terms with the [tiab] field for title/abstract. Combine concepts with AND, synonyms within a concept with OR. For Cochrane CENTRAL, mirror the MeSH/free-text structure but adapt to the CENTRAL syntax.

3A. PubMed

("MeSH Term A"[Mesh] OR "free text A1"[tiab] OR "free text A2"[tiab])
AND
("MeSH Term B"[Mesh] OR "free text B1"[tiab])
AND
(randomized controlled trial[pt] OR systematic review[pt])
AND
("2016"[dp] : "3000"[dp])

Include filters as appropriate:

  • Publication type: randomized controlled trial[pt], systematic review[pt], meta-analysis[pt], clinical trial[pt], observational study[pt].
  • Date range: ("YYYY"[dp] : "3000"[dp]).
  • Species (for human-only): humans[mh].
  • Language: english[la].

3B. Cochrane CENTRAL (CENTRAL/CDSR via Cochrane Library)

#1  MeSH descriptor: [Concept A] explode all trees
#2  (free text A1 OR free text A2):ti,ab,kw
#3  #1 OR #2
#4  MeSH descriptor: [Concept B] explode all trees
#5  (free text B1):ti,ab,kw
#6  #4 OR #5
#7  #3 AND #6
#8  #7 in Trials (CENTRAL) and Cochrane Reviews

3C. Guideline bodies (manual repository search)

For each, provide the canonical URL pattern and search terms:

BodyGuideline repositoryNotes
EFP (European Federation of Periodontology)https://www.efp.org/ — "Clinical Practice Guidelines" / "S3 Guidelines" sectionLook for S3-level guidelines (systematic review + GRADE + structured consensus).
AAP (American Academy of Periodontology)https://www.perio.org/ — "Clinical & Scientific Papers"Includes 2017 World Workshop staging/grading and treatment guidelines.
EAO (European Association for Osseointegration)https://www.eao.org/ — "Consensus Conferences"EAO Consensus Conference proceedings are the primary source.
ITI (International Team for Implantology)https://www.iti.org/ — "ITI Consensus"ITI Treatment Guides and Consensus Statements.
ADA (American Dental Association)https://www.ada.org/resources/research/science-and-research-institute/evidence-based-dental-researchADA Evidence-Based Dentistry guidelines.
NICE (UK)https://www.nice.org.uk/Where applicable (e.g., dental recall intervals, antimicrobial prophylaxis).
AAOMS (Oral & Maxillofacial Surgery)https://www.aaoms.org/For surgical guidance.

For each, write the search terms you would use in that body's site search.

3D. Trial registries

ClinicalTrials.gov:
   Search: [intervention] AND [population]
   URL: https://clinicaltrials.gov/search?term=...
   Filters: Phase, Status (Recruiting / Completed), Interventional, Results posted.

ISRCTN:
   https://www.isrctn.com/

EU Clinical Trials Register:
   https://www.clinicaltrialsregister.eu/

WHO ICTRP:
   https://trialsearch.who.int/

3E. Systematic review registry

PROSPERO:
   URL: https://www.crd.york.ac.uk/prospero/
   Search: [topic keywords]
   Use to find ongoing or completed systematic review protocols.

STEP 4: Live Retrieval (Only If Possible)

Execute the strategies only if Step 1 declared live retrieval = yes. For each source actually queried:

  • Report the exact query run.
  • Report the total result count returned.
  • Return the first 10–20 results as: Author Year. Title. Journal. DOI/PMID. — only what the source actually returned.
  • Flag any retrieval that failed (timeout, blocked, no access).

If live retrieval is no, skip this step. State explicitly: Live retrieval not performed — strategies above are for the user to execute.

Full text first

A search result is a title and an abstract. The next skill needs the paper.

  • When the runtime can run scripts, get the PDF of each paper you hand off with dental-paper-fetch before it is appraised. It uses legal open-access sources only.
  • Never appraise from the abstract when a free full text exists.
  • Otherwise ask the user for a PDF they may lawfully share.
  • Fill the "Full text obtained" line of the Retrieval Mode block. Before any paper is fetched, write "abstract only via search results". When papers differ, say so, for example "partial: 4 of 10 papers saved via dental-paper-fetch, 6 with no free copy".
  • Report each paper that did not download with the result line and link that dental-paper-fetch printed. Do not invent a link.

STEP 5: Retrieval Log

Produce a structured log the user (or clinical-evidence-reviewer) can consume:

## Retrieval Log

| Source | Query (verbatim) | Date | Results returned | Retrieval status |
|---|---|---|---|---|
| PubMed | [query] | YYYY-MM-DD | N hits | Searched / Not searched / Failed |
| Cochrane CENTRAL | [query] | YYYY-MM-DD | N hits | Searched / Not searched / Failed |
| EFP guidelines | [search terms] | YYYY-MM-DD | N relevant docs | Searched / Not searched / Failed |
| AAP | … | | | |
| EAO | … | | | |
| ITI | … | | | |
| ADA | … | | | |
| ClinicalTrials.gov | … | | | |
| PROSPERO | … | | | |

## Coverage statement
- Critical PICO concepts covered by search strategy: [yes/no per concept]
- Known limitations of this search: [single database / no grey literature / language filter / etc.]
- Suggested next steps if results are sparse: [add MeSH explosion, add hand-search of key journals, contact authors]

STEP 6: Hand-Off

After producing the retrieval log:

  • If the user wants the body of evidence graded: hand off to clinical-evidence-reviewer with the retrieval log attached.
  • If the user wants individual papers critiqued: hand off to research-critic per paper.
  • If the user wants numerical extraction, effect-size interpretation, SD / CI / MCID review, or statistical validity checked from retrieved papers: hand off to dental-statistical-forensics with the extracted outcomes and citation metadata.
  • If the user wants more sources: produce additional strategies for hand-searching specific journals (Journal of Clinical Periodontology, Clinical Oral Implants Research, International Journal of Oral & Maxillofacial Implants, Journal of Periodontology, Journal of Dental Research, etc.).

Output Format

## Retrieval Mode
[the block from Step 1]

# Evidence Retrieval: [Clinical Question]

## PICO
[the block from Step 2]

## Search Strategies

### PubMed
[strategy block]

### Cochrane CENTRAL
[strategy block]

### Guideline bodies
[per-body search terms]

### Trial registries
[ClinicalTrials.gov + others]

### PROSPERO
[search terms]

## Live Retrieval Results
[Only if Step 1 said yes. Otherwise: "Live retrieval not performed in this runtime."]

## Retrieval Log
[the log block from Step 5]

## Hand-Off
[next-step recommendation]

Example Prompts

  • "Build me a search strategy for immediate vs delayed implant placement in molar sites."
  • "What PubMed query would find the evidence on PRF in socket preservation?"
  • "Find the current EFP guideline on stage III periodontitis treatment."
  • "Search for ongoing trials on Er:YAG laser for peri-implantitis."
  • "Help me build a systematic-review search on hyaluronic acid for peri-implant mucositis."

Important Notes

  • No fabricated citations, ever. If you cannot verify a result, do not write it.
  • State retrieval status honestly. The user must be able to tell whether your output reflects an actual search or just a search plan.
  • Search strategies are reproducible — every Boolean and filter you list should be one the user can copy-paste into the source.
  • Re-running the same search on a different date may return different results. Always include the date in the retrieval log.

Methodology Review Date

Last methodology review: 2026-05-16

This skill must be re-reviewed when any of the following changes materially:

  • PubMed query syntax or MeSH structure.
  • Cochrane Library search interface.
  • URLs / repository structure for EFP, AAP, EAO, ITI, ADA, NICE, AAOMS.
  • ClinicalTrials.gov / PROSPERO interface.
  • Available retrieval tools in Claude Code / claude.ai / API runtimes.

Dated changes:

  • 2026-09-30 (full text): New subsection "Full text first" under Step 4 names dental-paper-fetch as the way to get the PDF of each paper that is handed off. The Retrieval Mode block has one new line, "Full text obtained: [yes / partial / abstract only] via [source]". Query syntax and repository URLs were not re-reviewed for this change.
  • 2026-09-30 (re-audit): The PICO field "Time horizon / publication window" is now "Publication window", and a separate "Follow-up horizon" line uses the follow-up bands of clinical-evidence-reviewer, so the two meanings of "time horizon" no longer share one field. Query syntax and repository URLs were not re-reviewed for this change.

Part of Dental AI Skills by Francisco Teixeira Barbosa

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.