Skip to content

zy839971925-zyy/Agent--deep-research-Workflow-Skills / modular/skills/deep-research

deep-research

Verified

Use 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.

references/research-to-report.md

View on GitHub
# From a research question to a report

Start with the reader's decision or question, including scope, date and what would change the answer. Allow an orientation search to discover terminology. For the deciding unknown, ask where an observable trace would exist. Search for that trace using entity names, versions, units or identifiers found in sources.

### Search and change course

An official manual establishes intended API behavior. A test on the pinned version can establish observed behavior. A study can support an effect only within its population, comparator and design. A practitioner account can reveal a failure worth testing. Choose accordingly. Open the relevant text, figure, code or release note. Follow secondary coverage to the original when the distinction changes confidence. If the leading account and its strongest rival predict different outcomes, search for the distinguishing observation. A failed search bounds that route; try a different source ecosystem, language or primary identifier when it matters.

For pivotal claims retain a short working note: claim; exact locator and scope; what it supports; lineage; counterevidence or remaining gap. Use the existing task state. For routine facts a direct source suffices. Group syndications and reports using the same dataset. Keep inaccessible sources labeled. Before compression preserve the exact locator of any result that reverses or narrows a conclusion. Reopen it when the summary cannot support the claim.

### Update the answer

When new evidence arrives, identify which premise changes. Recompute affected ranges, confidence or recommendations. Keep stable conclusions intact. Distinguish missing evidence from a negative result. Do not manufacture symmetry between a documented result and an unsupported alternative. If successive searches repeat an origin, try one genuinely discriminating route when available. End when the material decision is stable, or state exactly which unresolved condition can still reverse it.

### Write for use

Select the form by purpose: a direct answer for a bounded fact, a comparison table for matched dimensions, a narrative for a timeline, or a sourced report for a material investigation. Put the answer and its important condition early. Attach sources to the particular claims they support. Expose disagreements in terms of versions, regions, denominators, methods or time periods. Keep extrapolation visible. Preserve unknowns in the final recommendation and give the reader the implication. Inspect any decisive number, quote, attribution or clickable evidence locator before delivery.

For evaluation, use a fixed corpus to diagnose retrieval vs interpretation and a dated live-web run to assess current performance. When a decisive source is supplied directly after a failed search, a correct second answer is a clue about the bottleneck. Benchmark answers and graders must remain outside the working corpus. See [BrowseComp-Plus](https://texttron.github.io/BrowseComp-Plus/) and [ALCE](https://aclanthology.org/2023.emnlp-main.398/) for controlled retrieval and citation evaluation.
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: 115kUpdated Claude CodeCodexVerifiedInstall
add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
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: 115kUpdated Claude CodeCodexVerifiedInstall
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: 115kUpdated Claude CodeCodexVerifiedInstall