zy839971925-zyy/Agent--deep-research-Workflow-Skills / modular/skills/deep-research
deep-research
VerifiedUse when a scenario calls for real-world evidence, pre-build engineering/architecture research, or a linked hybrid; infer the route without requiring the user to label it. Keep the host-selected default model; infer proportionate workflow depth, honoring explicit depth requests such as Ultra.
| File | Kind | Size |
|---|---|---|
| LICENSE.txt | markdown | 1.0 KB |
| SKILL.md | skill md | 3.7 KB |
| references/comparative-synthesis.md | markdown | 3.2 KB |
| references/evidence-and-provenance.md | markdown | 7.3 KB |
| references/information-target-navigation.md | markdown | 7.4 KB |
| references/inquiry-and-research.md | markdown | 9.1 KB |
| references/multimodal-and-data.md | markdown | 3.4 KB |
| references/research-modes-and-methods.md | markdown | 17.3 KB |
| references/research-to-report.md | markdown | 3.0 KB |
| references/retrieval-and-observation.md | markdown | 14.2 KB |
| references/time-scenarios-and-forecasting.md | markdown | 3.6 KB |
| references/ultra-research.md | markdown | 9.7 KB |
references/comparative-synthesis.md
View on GitHub# Comparative evidence synthesis
**Trigger:** Several empirical studies, technical reports, models, or benchmarks are being compared and their apparent differences may reflect mismatched versions, populations, conditions, methods, or metrics.
**Do not load for:** A simple comparison with directly commensurable facts, or a question that only needs one source.
## Extract before comparing
Create a compact comparison card for each study or result. Capture only fields that affect interpretation:
| Field | Record |
| --- | --- |
| Identity and time | Exact paper/report, version/model, date, publication or correction status |
| Target and sample | Population, task, dataset, inclusion/exclusion, setting, and denominator |
| Design | Research design, intervention/comparator, protocol, time horizon, and relevant controls |
| System conditions | Model/checkpoint, prompt, tools, context, compute/budget, hardware, and evaluator where applicable |
| Outcome | Metric definition, unit, aggregation, effect/estimate, uncertainty interval or reported variance |
| Validity limits | Selection, confounding, measurement, benchmark contamination, missing data, external validity |
| Provenance | Primary source, reused data, sponsor/affiliations/conflicts, linked reports or source family |
Do not fill missing fields by guessing. Mark them unknown and decide whether the omission could change the comparison.
## Establish comparability
First state the estimand or question the comparison is meant to answer. Align task, population, version, protocol, metric, denominator, time window, and resource budget. A shared label such as “accuracy,” “efficacy,” or “benchmark score” is not enough if its operational definition differs.
- If the core target and measurement are sufficiently aligned, compare the results and retain their uncertainty.
- If only some conditions align, compare within those strata and name the boundary.
- If results answer different questions, present them side by side without ranking or pooling.
- Do not calculate a pooled effect or declare a winner unless the data and design justify that operation.
## Reconcile disagreements
When results disagree, test scope, model or intervention version, sampling, evaluator, prompt/tool budget, measurement, denominator, uncertainty, source lineage, and conflicts of interest before treating the disagreement as a true contradiction. Preserve negative findings and failed replications. Distinct sources that reuse one dataset or evaluation pipeline are not independent replication.
For model or benchmark comparisons, record whether the system had the same tools, search corpus, context, token/compute budget, and stopping rules. A performance difference under unequal conditions does not isolate a model effect.
## Use the card to improve the research frontier
After extracting the cards, identify which mismatch or missing datum could actually change the answer. If one can, name the exact source, table, appendix, or reproduction that could resolve it and pursue that next. Otherwise bound the comparison and stop. The card is an internal aid, not a mandatory table in every user-facing response.
[← Return to Deep Research](../SKILL.md)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |