research

スター17

フォーク1

更新日2026年6月15日 15:47

Force explicit research mode — Scope → Discover → Synthesize → Plan — with the approval gate before any implementation.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

state-alchemists

state-alchemists/zrb

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

/research

The user invoked /research with: $ARGUMENTS

Procedure:

Ensure core-research is activated for this session. If you can't recall its workflow from earlier, re-activate via ActivateSkill('core-research') — silent and auto-approved. It provides the canonical Scope → Discover → Synthesize → Plan workflow and the approval gate.
Treat the user's input as the research question. If the question is ambiguous or under-specified, complete the Scope step by asking once before discovering.
Apply the workflow. Do not modify code during Scope or Discovery. When running interactively, do not begin implementation until the user has explicitly approved the plan — this overrides the Working Loop's Frame step, which would otherwise work autonomously on directives. When running non-interactively (no user to approve), present the plan inline and then proceed.
Deliver per the workflow's output standards: direct answer (for questions), Action → Target → Verification steps (for plans), or compared options with rationale (for decisions). Cite sources: file:line for code, URLs for web.

このリポジトリの他の Skills

同じリポジトリ

core-journaling

state-alchemists/zrb

Ensure this skill is active before every journal write — new entry, edit, restructure, or activity log append. Provides the graph protocol (bidirectional links, indexes), the directory layout, and the activity log format that keep the journal consistent.

2026-06-2017

core-coding

state-alchemists/zrb

Activate when the turn's deliverable is source, test, code analysis, or config files — editing or creating code. Provides the Research → Strategy → Execution workflow, the complexity budget, and specialised deep-dive companions (testing, debug, refactor, review).

2026-06-2017

core-design

state-alchemists/zrb

Activate when the turn's deliverable is a design — system architecture, API contract, data model, component decomposition, or trade-off analysis. Provides the Constraints → Explore → Decide → Specify → Plan workflow for sound, well-reasoned designs.

2026-06-2017

core-research

state-alchemists/zrb

Activate when the turn's deliverable is findings, comparisons, recommendations, or an investigation-backed plan — answering questions, analyzing unfamiliar code, exploring new domains. Provides the systematic Scope → Discover → Synthesize → Plan workflow.

2026-06-2017

core-writing

state-alchemists/zrb

Activate when the turn's deliverable is substantial prose — docs, copy, proposals, feedback, commit/PR text, or UI strings. Provides copywriting frameworks, tone guidance, and editing workflow for clear, persuasive, polished output.

2026-06-1817

git-summary

state-alchemists/zrb

Prepare high-quality git commit messages or comprehensive Pull Request (PR) descriptions based on recent changes.

2026-06-1117

name	research
description	Force explicit research mode — Scope → Discover → Synthesize → Plan — with the approval gate before any implementation.
user-invocable	true
disable-model-invocation	true

/research

The user invoked /research with: $ARGUMENTS

Procedure:

Ensure core-research is activated for this session. If you can't recall its workflow from earlier, re-activate via ActivateSkill('core-research') — silent and auto-approved. It provides the canonical Scope → Discover → Synthesize → Plan workflow and the approval gate.
Treat the user's input as the research question. If the question is ambiguous or under-specified, complete the Scope step by asking once before discovering.
Apply the workflow. Do not modify code during Scope or Discovery. When running interactively, do not begin implementation until the user has explicitly approved the plan — this overrides the Working Loop's Frame step, which would otherwise work autonomously on directives. When running non-interactively (no user to approve), present the plan inline and then proceed.
Deliver per the workflow's output standards: direct answer (for questions), Action → Target → Verification steps (for plans), or compared options with rationale (for decisions). Cite sources: file:line for code, URLs for web.