Skip to content

hashgraph-online/awesome-codex-plugins / plugins/alpoxdev/hypercore/skills/hyper-research

hyper-research

Verified

[Hyper] Use this skill when the user asks for multi-source, cited research - fact-finding, comparisons, market or trend analysis, or evidence-backed recommendations across live web, official docs, GitHub, and local repo sources - saved as a markdown report. Do not use for one-source lookups, rewriting, or implementation.

rules/research-workflow.md

View on GitHub
# Research Workflow

Use this rule file when a research task starts. It owns framing, the fidelity record, the journal, the first collection wave, saving, and the closing briefing. Expansion, verification, and gates live in their own rule files.

## 1. Confirm The Request

- If the user gave no topic, ask for it in one question and stop before any collection.
- Pick the depth from the user's words (`SKILL.md`, trigger conditions); with no depth word, use the default mode.
- Mark the request date-sensitive when it says `latest`, `current`, `today`, `recent`, `this year`, or similar relative-date wording.

## 2. Frame The Question

Start from "what must be true if the user's premise is true?", not from "what does the first search say?". Write this block before any search, into `brief.md` (default, deep, parallel) or the report's scope section (`--quick`):

```text
Core question: <the actual information need, one sentence>
Axes: <axis - what to search, where, why>   (quick 1-2, default 3-4, deep 4-6; orthogonal, no shared angle)
Expected truths: <E1 - a statement that must hold if the premise is right>  (one per axis at least)
Entity: <the canonical name and first-party URL of anything the topic names, or "to settle in wave 1">
Codebase relevant: yes/no | External: yes/no | Execution checks likely: yes/no
Precision demand: <what a wrong claim costs here: low | normal | high (legal, medical, financial, security, public-facing)>
```

Expected truths are what the run checks reality against. Each one ends the run as `true`, `violated`, or `unknown`, and the report says which.

## 3. Record The Fidelity Fields

Three fields say what the report must stay faithful to. Infer each one first; ask only when the answer changes the outcome and nothing in the request, destination, or sources decides it.

| Field | Values | Default | Ask when |
|---|---|---|---|
| `as of` | latest at run start, a version, release, tag, commit, date, or quarter | latest at run start, with the run date | the topic is version-bound or time-sensitive and nothing names an anchor |
| `source floor` | `standard`, `primary only`, `pointed only: <ids/URLs/paths>` | `standard` | a restricting phrase ("official docs only") sits beside a question whose answer lives outside that floor |
| `derived numbers` | `allowed`, `none` | `allowed` | the request both asks for comparisons or growth and says "only the sources' own numbers" |

Record each field with `answered_by: request | default | user`. Put at most one question per field in a single message, default option first, and start collection without waiting; a late answer is folded in before the report is written. Pointed-at sources with no restricting word are seeds, not a floor.

## 4. Open The Journal

For default, deep, and parallel runs, create `.hyper/research/[NN].slug/` beside the future report and keep the files described in `references/journal-format.md`. `--quick` keeps its ledger and query log inside the report.

- The leader writes every journal file; lanes never do.
- Each source, quote, number, and lead is written the moment it lands. The journal is the recovery point after context loss: after a compaction, re-read `brief.md` and the journal before doing anything else, then resume from the open wave.

## 5. Plan Channels And Queries

Read `references/channel-selection.md`. For each axis, write the distinct queries and the channel each targets into the query log before running them. Search in English first; add a local-language sweep when the topic is local or the user names a source language.

## 6. Run The First Wave

- Launch every axis in the first wave at once - as lanes when the parallel modifier applies (`rules/parallel-research.md`), otherwise as consecutive searches by the leader. Do not "start with one axis and see".
- The first wave settles the entity when one is unresolved: canonical name, first-party URL or account, whether it exists in the claimed category, and a confidence line. An unresolved entity never becomes a premise in a later query or lane prompt.
- Open every candidate before counting it. A result snippet is a lead; the opened artifact is the source.
- For each opened source, add a ledger row with publisher, dates, channel, grade, and role, then note the exact quote or line the claim rests on.
- When a channel is unavailable, say so in the query log and use the next-best channel rather than inventing a tool.

Then hand over to `rules/expansion-and-convergence.md`.

## 7. Save The Report

Save the report to `.hyper/research/[NN].slug.md`, where `NN` is the next zero-padded sequence number in `.hyper/research/` and `slug` is a short ASCII summary of the topic. The journal directory uses the same `[NN].slug` stem. Create the report file with its section headings as soon as the brief exists, so an interrupted run leaves a partial report on disk instead of nothing.

When the report file is created, its first five lines carry `STATUS: draft`. That line stays while any gate is open and is removed only after the last gate runs. A run that finds `STATUS: draft` on the existing report resumes from the journal instead of starting a new report file.

The journal brief records the deliverable promise next to the fidelity fields: the format (`md` by default), the path, and the `answered_by` value that chose the format (`user`, `default`, or `request`). The promised format and path are what the saved report is measured against in `rules/validation.md`.

## 8. Closing Briefing

The final chat message carries, in this order:

- the headline conclusion in one to three sentences
- the saved report path, and the journal path when one exists
- an effort line: sources reviewed and cited, distinct domains, waves, excursions, and execution checks, read off the journal rather than recalled
- after the gates run, `bash scripts/report-state.sh finish --state <journal>/report-state.json --report <report.md> --ledger <journal>/sources.md` writes `outcome.json` into the journal; `report-state.sh briefing --state <journal>/report-state.json` prints the sources, domains, repairs, gate results, and residual defects counted from that file. Use those printed numbers in the effort line, not recalled ones.
- the run state (`converged` or `exhausted`) and the main caveat or open gap when confidence is limited

## Sources

> Checked 2026-11-03.

| Claim | Source |
|---|---|
| Framing from expected truths, the analysis block, entity disambiguation in the first wave, the real-time journal, the early report skeleton, and the closing effort line, adapted from the `ulw-research` skill | <https://github.com/code-yeongyu/oh-my-openagent/tree/0d80470e4201925727e88afbafcdf24d7801a588> |
| The three fidelity fields, reduced from the upstream seven, with inferred defaults and `answered_by` | same source, reduced by this package |
| A snippet is a lead and never primary evidence; reviewed and cited are counted apart | `instructions/sourcing/references/source-ledger.md` |
| Keep the authoritative plan outside the context window and re-read it after compaction | `instructions/context-engineering/CONTEXT_ENGINEERING.md` |

### Evidence grade

`LOCAL` - this file states this package's own procedure, adapted from the upstream skill named above. The axis counts and the fidelity field set are this package's choices.
rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content.Stars: 197kUpdated Claude CodeCodexVerifiedInstall
claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).Stars: 197kUpdated Claude CodeCodexUnverifiedInstall
doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.Stars: 197kUpdated Claude CodeCodexVerifiedInstall