Skip to main content

risk-compliance-super-intelligence

Risk & Compliance (GRC) Super Intelligence Team. Named real-world personas across 7 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md.

来源信息

仓库
coco-research/coco
最近来源活动
2026年9月15日 15:30
检测到的 SKILL.md 语言
英语
星标
466
分支
12

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

文件资源管理器
51 个文件

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
risk-compliance-super-intelligence
description
Risk & Compliance (GRC) Super Intelligence Team. Named real-world personas across 7 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md.
team_id
risk-compliance-super-intelligence
team_name
Risk & Compliance (GRC) Super Intelligence Team
last_updated
2026-06-01
schema_version
1.0
# Risk & Compliance (GRC) Super Intelligence Team Risk & Compliance (GRC) Super Intelligence Team. Named real-world personas across 7 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md. **28 native personas** (12 cross-listed) across **7 cells**. Real public figures rendered as illustrative composites — see `superintelligence/DISCLAIMER.md`. Built local-first + Claude-research, validator-gated (real cited URLs, no fabrication). ## Cells | Cell | Personas | Focus | |---|---|---| | `privacy-data-protection` | 6 | Privacy law, data protection, GDPR/CCPA, privacy-by-design. | | `ai-governance-responsible-ai` | 8 | AI governance, responsible AI, algorithmic accountability, EU AI Act / NIST RMF. | | `regulatory-tech-policy` | 6 | Tech regulation, antitrust, platform policy, algorithmic-accountability journalism/law. | | `security-governance` | 5 | Security governance, disclosure, threat policy (cross-listed from Engineering). | | `erm-grc-leadership` | 6 | Enterprise risk management, GRC frameworks, internal audit, three-lines model. | | `financial-crime-aml` | 4 | AML/KYC, sanctions, financial-crime compliance. | | `model-systemic-risk` | 5 | Model risk, validation (SR 11-7), systemic/financial-stability risk (cross-listed from Trading). | ## Command surface - **Orchestrate (smart panel):** `/SI-GRC-Orchestrate "<prompt>"` → picks 16-32 personas + approval gate - **Action verbs:** `/SI-GRC-Decide` · `-Tradeoff` · `-Pre-Mortem` · `-Review` · `-Stress-Test` · `-Plan` · `-Design` · `-Analyse` · `-Vote` · `-Debug` · `-Defend` · `-Roast` · `-Post-Mortem` · `-Re-Analyse` · `-Full-Cycle` - **Identity:** `/SI-GRC-Ask <slug> "<q>"` · `-Huddle <cell> "<topic>"` · `-Meeting "<prompt>"` · `-Read <slug>` - **Roster/maintenance:** `/SI-GRC-Recruit` · `-Refresh` · `-VoiceCheck` · `-Verify` ## Cross-team For multi-domain decisions, the top-level `/SI-Orchestrate` (+ `/SI-<Verb>`) routes ACROSS teams via `superintelligence/scripts/meta_select.py` (local nomic-embed). This team auto-joins that panel. ## Source of truth `registry.json` (generated from persona frontmatter) + `cells/*.md` + `roster.json`. Regenerate: `python3 risk-compliance/scripts/build_registry.py && python3 risk-compliance/scripts/build_cells.py`.
在 GitHub 查看