Deep evidence-first research with broad discovery, verification, and traceable citations. Prefer invoking via research-workflow. TRIGGER when (MANDATORY — you MUST invoke this skill, no exceptions): user message contains ANY of these keywords or synonyms — 调研/研究/对比/综述/文献/证据/机制/根因/为什么/可行性/路线图/分析/探索, or research/investigate/compare/survey/literature/evidence/mechanism/root-cause/why/feasibility/roadmap/analyze/explore — or asks to verify claims, analyze tradeoffs, scope a new topic, or conduct literature review. Also use this skill as the default gateway for external search. Skipping when keywords match is a routing violation. DO NOT TRIGGER when: user asks for paper-writing output (use paper-writing), experiment launch (use experiment-execution), or plan-only without evidence (use research-plan).
Execute AI/ML experiments locally or remotely with environment, runtime, and logging controls. Prefer invoking via research-workflow. TRIGGER when: user asks to run/launch/start/resume/monitor a training job, evaluation, or benchmark, or a plan is ready for execution, or experiment needs rerun/recovery. DO NOT TRIGGER when: research investigation only (use deep-research), planning without execution (use research-plan), or env setup without launch (use project-context).
Manage long-term AI R&D memory: retrieval, writeback, promotion, and shared export. TRIGGER when: run bootstrap, each new user turn, each execution batch, significant failure, replan, high-resource action, long-action resume, final report handoff, or compaction markers detected (Compact/压缩/Summary). DO NOT TRIGGER when: the exact same retrieval was just performed, freshness is still valid, and no new objective/stage/error signal appeared.
PRIMARY ORCHESTRATOR — trigger this skill FIRST for any non-trivial AI R&D task. Coordinates run-governor, memory-manager, deep-research, research-plan, project-context, experiment-execution, human-checkpoint, and paper-writing. TRIGGER FIRST when: any non-trivial research task begins (analysis, debugging, reproduction, planning, evaluation, training), or context compaction detected (Compact/压缩/Summary). Other skills should be invoked FROM WITHIN this workflow. DO NOT TRIGGER when: trivial single-command tasks ("show file", "fix typo"), or pure conversational queries.
Govern run-level execution policy: mode selection, durable run tracking, long-action watch/resume policy, stage reporting, and safety allowances. TRIGGER when: starting a non-trivial research task (set mode + run_id), switching local/remote target, creating a new run, or mode-aware policy decisions needed. DO NOT TRIGGER when: trivial single-step tasks, pure info queries, or run already initialized with no policy change needed.
Escalate critical decisions to a human with options, tradeoffs, and recommendation. TRIGGER when: major safety risk, high irreversible impact, large GPU spend, destructive data ops, shared-memory publication, or hard blockers requiring human approval. DO NOT TRIGGER when: routine confirmations handled by run-governor, or low-stakes reversible actions.
Write CS/AI papers with progressive disclosure. Prefer invoking via research-workflow. TRIGGER when: user explicitly asks to draft/write/revise a paper, paper section (abstract, intro, related work, method, experiments, conclusion), rebuttal, or LaTeX content. DO NOT TRIGGER when: user mentions papers but needs research/investigation (use deep-research), experiment planning (use research-plan), or experiment execution (use experiment-execution).
Initialize and maintain per-project runtime context (env, secrets, snapshots). Prefer invoking via research-workflow. TRIGGER when: new run needs env setup, preflight before experiment/eval, runtime fields missing (paths, API keys, GPU config, proxy), run snapshot needed, or shared-memory needs project config. DO NOT TRIGGER when: pure research (use deep-research), plan-only (use research-plan), or env already fully resolved.