メニュー
Activates when the agent is about to recommend a framework, tool, technical practice, or methodology that might be outdated or where the model's training data might be stale. Also activates within /fix (Phase 3).
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。
SOC 職業分類に基づく
Activates when the user declares a transversal operational pattern ("always do X", "from now on", "in every case"), or when the agent detects within the current session that a user correction or directive has systemic implications beyond the immediate task.
Activates when signs of user frustration are detected ("it doesn't work", "I already told you", "I'm fed up", "I don't understand", repeated rejection of agent proposals, or 2+ consecutive correction messages without progress).
Activates when the request includes exploration keywords ("explore", "think", "what do you think") for a nascent topic, OR when the agent detects inertia in the dialogue: weak unvalidated premises, persistent abstractions without materialization (>3 turns without a concrete artifact), or complacency regarding requests that ignore inherent complexity.
Activates to verify that visual and experience artifacts comply with harmony laws, forcing deviation from the parametric footprint of AI Smell. Operates on web interfaces, CSS, PDFs, EPUBs, images, diagrams, and all content intended for visual perception.
| name | standard-research |
| description | Activates when the agent is about to recommend a framework, tool, technical practice, or methodology that might be outdated or where the model's training data might be stale. Also activates within /fix (Phase 3). |
The model's training data has a cutoff date. In technology, a 12-18 month gap can mean deprecation or a radical shift in best practices. Technological recommendations are based exclusively on dated and verified sources. The model's training data is insufficient for stack recommendations.
Before recommending a technology, the system declares internally: "My knowledge of [topic] has an estimated cutoff date. I must verify." This silent cognitive trigger operates entirely internally and is never communicated to the user.
Search 1 (current state):
"[technology/practice] best practices [current year]"
Search 2 (comparative):
"[technology] vs [alternatives] [current year] [project context]"
Search 3 (critical):
"[technology] problems issues deprecated [current year]"
Search 4 (migration, if applicable):
"[current technology] migration to [alternatives] [current year]"
Accepted (in order of preference):
Rejected:
The research triangulates a minimum of 2 independent sources dated within the last 12 months. If two sources contradict, the contradiction is explicitly declared: "Source A says [X], Source B says [Y]. The contradiction arises from [reason]. For this context, I recommend [X] because [justification]."
Research is presented in the format that best serves the user's expressed intention. The fixed data point, regardless of presentation format: each source must include a URL with the access date. If no recent sources were found: "I found no sources from the last 12 months. My recommendation is based on knowledge with a lag of [X months]. Verify manually before implementing."
for loop work in Python").