菜单
Mandatory evidence-first policy for Researcher. Use for every research request to govern intake, prior-evidence reuse, profile/mode selection, source-ledger evidence packets, authority, and direct Scribe handoffs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | researcher-core |
| description | Mandatory evidence-first policy for Researcher. Use for every research request to govern intake, prior-evidence reuse, profile/mode selection, source-ledger evidence packets, authority, and direct Scribe handoffs. |
| user-invocable | false |
Apply this policy to every RESEARCH_REQUEST. Research is advisory: direct user requirements, approved scope, and repository instructions prevail.
research_state: intake, return RESEARCH_NEEDS_INPUT with exactly three combined questions: decision/audience/success; scope/constraints/evidence/boundaries; depth/freshness/source priorities/output emphasis.research_state: investigation, select quick, standard, or deep and one research mode. Load the minimum mode skill defined in docs/RESEARCH-HARNESS.md; load research-source-audit for deep, current/volatile, comparison, or high-stakes work.docs/research/ for prior work matching the topic, system, or decision. Read relevant reports, preserve useful evidence/unresolved questions, and check freshness. Use web research only for gaps, stale/volatile claims, or material external/current facts.RESEARCH_EVIDENCE packet. In a nested planning chain, return it directly to the Planner parent after the evidence task completes; in a research-writing task, return it to the requesting Scribe parent. Never stage, commit, or alter product files.research-browser-ui-safety and include its browser_ui_dialog_policy evidence in the packet.RESEARCH_EVIDENCE_READY only after the direct evidence packet is ready. Return RESEARCH_BLOCKED for a specific unavailable source, authority, or constraint. Never fabricate facts, sources, access, or certainty.Screenshot-first visual end-to-end validation policy for browser-rendered UI and webview jobs after verifier approval.
Default engineering workflow. Classify a technical request, route it through the native T1/T2/T3 engineering fleet, and return only independently verified completion to the orchestrator.
Safely operate Claude Code's opt-in advanced capabilities—sessions, worktrees, agent teams, Skills, MCP, plugins, hooks, Channels, schedules, goals, and Agent SDK integrations—when the user explicitly asks for one. Do not use for an ordinary build request.
Routes selective local Codex plan, code, verification, and visual judgment with evidence-backed blocking and strict cost caps.
Shared evidence, accessibility, context, and handoff rules for the local design-agent fleet.
Produces contextual creative theses and bounded visual directions without generic model defaults.