Manuscript Srma

Evidence-synthesis manuscripts

Draft only from supplied or verifiable evidence. Treat this skill as a workflow and house-style layer, not as a substitute for a protocol, statistician, reporting guideline, or a journal's current instructions.

Verify changing requirements first

Before asserting compliance or applying an exact requirement:

1. Open the current official author instructions for the exact journal and article type.
2. Open the current official reporting-guideline checklist or extension appropriate to the review.
3. Record the pages and access date in working notes.
4. Prefer the current official source over every bundled snapshot. If browsing is unavailable, label volatile requirements unverified, insert a placeholder, and do not claim submission readiness.

Verify at least the abstract headings and limit, main-text and display-item limits, supplement rules, required checklists or flow diagrams, registration language, title/subtitle format, contribution and disclosure statements, data/code policies, and any generative-AI disclosure. For reporting guidance, verify the applicable PRISMA extension, SWiM or other design-specific guideline, and current GRADE guidance. Never assume that a requirement for one journal in a family applies to another.

Establish the project state

Inspect the request and supplied artifacts before asking questions. Record:

• mode: write, edit, or evaluate;
• synthesis: systematic review without pooling, pairwise meta-analysis, network meta-analysis, or both pairwise and network;
• report stage: initial report or update;
• target journal and article type;
• requested deliverable and sections;
• available evidence, tables, figures, protocol, registry record, analysis output, and prior report;
• living status and interactive-platform status.

Treat living and interactive as separate toggles. Their initial state is unknown and operationally off: omit living-search, automation, and website language unless the user or source material confirms it. Do not infer either toggle from the topic, lab, repository, or prior publications.

Ask one compact question only when an unresolved choice materially changes the requested output. Otherwise continue with conservative, stated assumptions. For a small section request, draft that section rather than delaying for a full intake. Never repeat a question already answered by the files or conversation.

Use these applicability labels in the bundled references:

• [ALL]: every evidence synthesis;
• [SR-only]: synthesis without statistical pooling;
• [MA]: pairwise or network meta-analysis;
• [NMA]: network meta-analysis only;
• [LIVING] and [INTERACTIVE]: only after the corresponding toggle is confirmed on.

Enforce evidence integrity

• Reconcile every count and estimate across abstract, text, tables, figures, and supplement.
• Do not invent data, citations, dates, analyses, registry identifiers, software versions, author roles, or journal rules.
• Mark missing material with ⟦PLACEHOLDER — concise description of the needed input⟧. Use one complete token per gap; never hide gaps in fluent prose.
• Preserve effect direction, reference group, outcome time point, scale, analysis population, and uncertainty interval.
• Distinguish no evidence, no statistically detectable difference, imprecision, and evidence of no important effect.
• Report certainty by outcome or comparison, not as a single blanket rating for the review.
• Interpret network rankings only with effect sizes, uncertainty, certainty, and the direct/indirect evidence; never turn a ranking into a head-to-head claim.
• Describe registration, librarian involvement, automation, dual review, patient involvement, data/code sharing, or platform peer review only when supported.
• End newly drafted material with an Outstanding placeholders list, or state None after checking.

Run the requested mode

Write

1. Build a source-of-truth evidence inventory from the supplied artifacts.
2. Read sections.md for the requested sections.
3. Read methods-stats.md for Methods, synthesis, estimates, certainty, or statistical interpretation.
4. Read boilerplate.md only for relevant reusable language; replace every placeholder and delete inapplicable clauses.
5. Read exactly one journal adapter when a target is named: European Urology, JAMA, or Mayo Clinic Proceedings. Verify it against the live official instructions.
6. Draft only the requested deliverable, using placeholders for unsupported content.
7. Reconcile numbers and run the placeholder script on the deliverable.

Edit

1. Infer the project state from the draft and ask only for a decision that cannot be safely inferred.
2. Read sections.md, plus methods-stats.md when statistical content is in scope.
3. Preserve the author's substantive meaning and citation keys. Tighten structure, calibrate claims, and flag rather than silently repair unverifiable facts.
4. Verify the current journal and reporting requirements before describing the revision as compliant.
5. Return the revised text plus a concise list of substantive changes and unresolved placeholders.

Evaluate

1. Determine the synthesis type from the manuscript, protocol, or Methods; mark it unresolved if the sources conflict.
2. Read and apply evaluation-rubric.md, using only applicable tags.
3. Run scripts/find_placeholders.py on text artifacts and, when useful, scripts/extract_estimates.py to create a reconciliation inventory.
4. Rate each applicable rubric item Pass, Partial, Fail, or Not assessable, with evidence by page, section, table, or line.
5. Treat critical failures as blockers. End with a prioritized fix list and a separate list of inputs needed from the authors.

Use bundled resources

Resolve SKILL_DIR to the absolute directory containing this SKILL.md before invoking bundled files; do not assume the user's current working directory is the skill directory. Use absolute paths derived from SKILL_DIR for every script or asset operation.

Run the scripts with Python 3:

SKILL_DIR="/absolute/path/to/manuscript-srma"
python3 "$SKILL_DIR/scripts/find_placeholders.py" MANUSCRIPT_OR_DIRECTORY
python3 "$SKILL_DIR/scripts/extract_estimates.py" MANUSCRIPT_OR_DIRECTORY --format tsv

The placeholder script returns status 1 when it finds gaps (use --exit-zero for a nonfailing inventory) and status 2 on input errors. The estimate script supports common text formats and DOCX; its output is a reconciliation aid, not proof that values are correct.

For a manuscript file format such as DOCX or PDF, also use the relevant document or PDF workflow to preserve layout and verify the rendered deliverable. Keep scientific evaluation separate from file-format mechanics.