PhysioKeys

View saved version →

---
name: compare-interventions
description: Use when a user asks to compare rehabilitation interventions, exercise approaches or treatment alternatives using published research, including which approach is better and where results differ.
---

# Compare interventions

Read [the evidence and tool contract](../../references/evidence-workflows.md) first.

1. Establish the condition/population, alternatives and outcomes. Use a supplied comparison directly; clarify only a difference that would change which evidence applies.
2. Choose depth from the shared [adaptive search and reading budgets](../../references/evidence-workflows.md#adaptive-search-and-reading-budgets). If the comparison is limited to supplied or selected papers, skip discovery and read that set. For discovery, start with one search naming both alternatives and the condition, using the [source-selection policy](../../references/evidence-workflows.md#source-selection-and-pubmed-approval) and the live schema's supported result maximum. Prefer relevant direct comparisons. Refine to fill an identified comparator, outcome or follow-up gap within the whole-task budget. Read chosen papers in batches using the shared [exact-article full-text fallback](../../references/evidence-workflows.md#exact-article-full-text-fallback) for their access and provenance. A thorough comparison can require more than five papers. Use the shared source-specific ledger for larger comparisons.
3. Build a readable comparison table, combining columns where helpful, that preserves source/citation, population and sample size if reported, intervention and comparator, outcome/measure, follow-up, reported result/uncertainty, and evidence access. Use “not reported in available text” for missing facts instead of invented values. If no relevant paper was returned, give a short explanation of the search limits rather than an empty table.
4. Separate **direct comparative findings** from **indirect evidence**. Two trials against different controls do not establish which intervention is superior. Compare like outcomes and timepoints; do not pool or rank incompatible populations, scales or follow-up periods. A statistically nonsignificant result does not establish equivalence.
5. Explain what the evidence supports for the user's general research question, what prevents a fair comparison, and what a useful head-to-head study would need to resolve. Do not manufacture a winner, numerical effect, formal certainty grade or patient-specific recommendation.
6. State source coverage, search-call count, unique candidates screened, papers actually read/included and access limitations, sized to the request. Report any host fallback attempts separately. Do not turn screened snippets into claims of completed reading. If asked to save cited papers, use only verified PMIDs and the existing save tool, report each actual outcome, and identify UUID-only sources that cannot be saved through that tool. Do not save merely because a paper was cited.

Example: eccentric loading versus heavy slow resistance for Achilles tendinopathy. A 12-week pain trial against waitlist and a 6-week function trial against usual care belong in separate rows; their results cannot determine a treatment winner.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: evidence-brief
description: Use when a user asks for a focused evidence summary or research brief about physical therapy, physiotherapy or rehabilitation, including effectiveness, dosage studied, prognosis or assessment.
---

# Evidence brief

Read [the evidence and tool contract](../../references/evidence-workflows.md) first. For a comparison use [compare-interventions](../compare-interventions/SKILL.md); for appraisal of a particular paper use [paper-appraisal](../paper-appraisal/SKILL.md).

1. Identify the population, intervention/exposure, comparator if relevant, and outcome from the user's question. Ask one targeted clarification only if its absence would materially change the search. Preserve a clear question already supplied; never demand a formal PICO form.
2. If the brief is limited to supplied or selected papers, skip discovery and read that set. Otherwise choose depth using the shared [adaptive search and reading budgets](../../references/evidence-workflows.md#adaptive-search-and-reading-budgets), then start one focused `search` using the [source-selection policy](../../references/evidence-workflows.md#source-selection-and-pubmed-approval). For treatment questions, suitable reviews and RCTs are useful starting points; prognosis and measurement questions may need different designs. Use the live schema's supported result maximum and exact filter values. A clear deep-review request needs no redundant permission question within its announced budget.
3. Screen and deduplicate candidates, then read the chosen papers before claims. Follow the shared [exact-article full-text fallback](../../references/evidence-workflows.md#exact-article-full-text-fallback) for their access and provenance. Adapt the reading effort to the requested breadth; use the shared evidence ledger and coverage checkpoints for deeper reviews. Preserve distinct and conflicting findings, avoid double-counting trials included in reviews, and leave unreported facts unknown.
4. Deliver a concise answer in this order:
   - **Answer:** what the retrieved evidence supports, with appropriate uncertainty.
   - **Evidence:** for each source, population/design, intervention/comparator, relevant outcome and follow-up, reported effect/uncertainty when available, and citation/link. A compact table is useful for several papers.
   - **Limits and applicability:** differences between study populations and the question, missing methods/harms, and access depth. Do not infer a precise exercise prescription or certainty grade from study labels or duration alone.
   - **Scope:** source(s), design/year restrictions or relaxation when applicable, search calls, unique candidates screened and papers actually read/included, and remaining coverage gaps. Report any host fallback attempts separately. Use a concise scope note for a brief and fuller accounting for a deeper review; never imply that all returned candidates were read.

Example request: “Give me a brief on exercise for knee osteoarthritis.” The answer should distinguish review findings from a single trial, identify the outcome and follow-up, and state when full text or effect sizes were unavailable. A user asking for speed still needs those limits; keep them short rather than inventing precision.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: paper-appraisal
description: Use when a user asks for a journal-club review, critical appraisal or methodological assessment of a specific physical therapy or rehabilitation research paper identified by URL, title, PMID, DOI, attachment or a returned article record.
---

# Paper appraisal

Read [the evidence and tool contract](../../references/evidence-workflows.md) first.

1. Resolve and read the exact paper using the shared [exact-article full-text fallback](../../references/evidence-workflows.md#exact-article-full-text-fallback). Fetch a verified PMID or returned article UUID directly when available. Apply the shared [source-selection policy](../../references/evidence-workflows.md#source-selection-and-pubmed-approval) if a metered identifier lookup is authorized and needed. Clarify ambiguous matches instead of substituting a similar paper. Verify design from the available text.
2. State what was accessible and which source supplied it. Preserve the PhysioKeys access flags and separately label any host-read text, version and limitations. If no verified article text was accessible, stop short of an appraisal based on a snippet.
3. Produce these sections, sized to the paper and available evidence:
   - **Question and design:** population, research question, design and comparator.
   - **Methods and potential bias:** recruitment, allocation/confounding, blinding where relevant, missing data, outcome selection and analysis. For reviews, examine search/selection, included designs, heterogeneity and overlapping studies when reported. For diagnostic/measurement or prognosis studies, use their appropriate validity questions. Mark unavailable details as unknown; distinguish lack of reporting from demonstrated bias.
   - **Findings:** actual outcome measures, effect sizes, uncertainty and timepoints when supplied; separate statistical significance from practical importance. Do not invent MCIDs or assume a nonsignificant result proves equivalence.
   - **Applicability and limits:** how the study population, setting, intervention and follow-up relate to the research question; address harms only to the extent reported.
   - **Journal-club discussion:** two or three specific questions tied to the actual reporting gaps or findings.
   - **Citation and access note:** verified citation and identifier, the source URL and version actually read, and the limits on this appraisal.
4. Apply the shared [adaptive search and reading budgets](../../references/evidence-workflows.md#adaptive-search-and-reading-budgets). One identified paper needs no discovery search. Report any host fallback attempts separately. If the user supplies a larger paper set, work in batches and account for every requested ID, including duplicates, failed/unavailable records and any pending work. Keep each appraisal's facts and limitations tied to its own record.
5. Do not invent an official PEDro score, risk-of-bias result or GRADE certainty. If a fetched source reports a rating, attribute it and its scope explicitly. One paper and missing methods do not support an outcome-level body-of-evidence rating. Never assume unreported methods were performed correctly.

Example: if an RCT abstract omits allocation concealment, attrition counts, confidence intervals and harms, those fields remain unknown. Explain what full-text details would resolve them; do not fill them using typical trial practice.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: setup
description: Use when a user installs PhysioKeys, runs its setup workflow, asks how to connect or get started, or needs help recovering its AI-assistant connection.
---

# Set up PhysioKeys

Read [the evidence and tool contract](../../references/evidence-workflows.md) and [the library and chat guide](../../references/library-and-chat.md) first. Setup helps users connect and choose a useful task without spending a search just to demonstrate installation.

1. Preserve the current conversation's task. In a fresh conversation, briefly explain that PhysioKeys searches rehabilitation research, reads abstracts and available licensed full text, supplies citations, and can save requested PMIDs to the user's private library. State that requests must be de-identified general research questions. Do not solicit clinical records or identifiers.
2. If the PhysioKeys account tool is available, call `physiokeys_account_status` once to verify access. If unavailable, unauthorized or expired, direct the user to the host's PhysioKeys Connect/Reconnect control, complete OAuth in its secure flow, then retry only after the user has connected. The existing server URL is `https://app.physiokeys.com/api/mcp` when a compatible client's custom-MCP form asks for it. Do not assume host menu labels or create a second plugin. Never ask for passwords, access tokens or codes in chat, and do not claim a connection succeeded without a successful tool result.
3. Explain the actual returned allowance when known: each research search and continuation uses one normal PhysioKeys search. Opening saved papers, filtering their titles, picking or attaching a citation, reading article details and checking account status do not consume search allowance; saving counts toward saved-article capacity. Rate limits still apply. Explain the [source-selection policy](../../references/evidence-workflows.md#source-selection-and-pubmed-approval). Research effort follows the question through the [adaptive budgets](../../references/evidence-workflows.md#adaptive-search-and-reading-budgets): quick briefs stay small, while explicit deeper reviews can screen and read more within stated bounds. Result pages support up to 50 only where the live connection advertises it; respect smaller schemas and user/remaining allowances. Do not promise unlimited use, quote invented plan amounts, initiate checkout or promote an upgrade.
4. If searches are exhausted, explain that new searches cannot run. A provided PMID or relevant saved articles can still be read with the free tools. Do not claim those sources exist until returned. If the status call fails, report allowance as unknown.
5. Offer a free first action: open **Saved papers** in **Research Library** when the connected host exposes it. Explain **Filter saved papers** versus **Search research · uses 1 search**. Where available, **Papers in this chat** lets the user pick saved papers and attach their citations to the current conversation; supported desktop composers also offer a saved-paper picker through `@PhysioKeys`. Appraisal/comparison controls create an editable question for the user to review and send. Follow the library and chat guide for capability checks, evidence-access labels and fallbacks. Do not promise these surfaces from the package version alone, guess a tool name, open another chat, attach library contents, save a paper, or run a demonstration search automatically. With only the original tools available, offer to list saved papers on request and fetch a selected identifier.
6. When there is already a clear, authorized research task, continue with the appropriate skill rather than asking the user to repeat it. A saved-paper or supplied-ID task can use free reads even with no search allowance. The research skills follow the [exact-article full-text fallback](../../references/evidence-workflows.md#exact-article-full-text-fallback) if more text is needed and host access is allowed:
   - [evidence-brief](../evidence-bri

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.