coco-research/coco/ai

AI Super Intelligence Team — SKILL entry

Your AI research + engineering brain trust and decision-making partner. 59 named personas drawn from the world's frontier model labs (US, UK, China, France), top AI universities, the open-source ecosystem, the independent-evaluation and AI-governance ecosystem, and product-history archetypes. Reusable across any CoCo-routed prompt — invoked by /superintelligenceTeam-* slash commands. The team's primary purpose is to help take decisions, not just review work after the fact — every cell carries a decision-relevant lens that convene synthesis explicitly draws on.

Status: All three phases complete.

- Phase 1: 49 personas across 8 cells (baseline roster + founding v2 panelists).

- Phase 2: 10 priority additions from the EXPANSION.md gap analysis (roster grew to 59).

- Phase 3: 22 slash commands installed at ~/.claude/commands/superintelligenceTeam*.md — orchestrator-first action surface plus explicit-override identity surface. See Slash Commands section below.

This file is the user-facing entry point for the team. The machine source of truth is `registry.json`, regenerated from persona frontmatter by python3 superintelligenceTeam/scripts/build_registry.py.

What this team is for

When a CoCo prompt is high-stakes enough to deserve a parallel review panel, the AI Super Intelligence Team plays the role of named external voices. Instead of "the panel said," every claim is attributed to a specific researcher with documented stances. This lets convene synthesis:

• Surface real disagreement (Hinton vs LeCun on existential risk; Karpathy vs Schulman on RL).
• Anchor stances to citable evidence (every public_stance in every persona has an evidence_url).
• Stay honest about who actually drove which decision (lead-driver vs validator vs specialist vs swing).

The 5-cell × 4-persona founding Memory v2 panel from May 26-27, 2026 was the first consumer. This team supersedes that roster: five of those panelists (Karpathy, Wei, LeCun, Tri Dao, Hyung Won Chung) continue here with their v2_panel_attribution records preserved. The rest of the founding v2 panel (memory / cloud / security / privacy personas) belong to future Super Intelligence Teams (Memory Systems, Cloud, Data, Compliance) and are not part of this AI-focused roster.

Cells (8)

Personas (59)

Listed by cell. Each has a YAML-frontmatter profile under personas/<slug>.md plus a research dump under research/<slug>/. Bold marks Phase 2 additions (built 2026-05-28 from EXPANSION.md).

Frontier Labs Research (6): ilya-sutskever · dario-amodei · demis-hassabis · jakub-pachocki · [liang-wenfeng](personas/liang-wenfeng.md) · [arthur-mensch](personas/arthur-mensch.md)

Applied AI Leadership (7): sam-altman · mira-murati · greg-brockman · aravind-srinivas · aidan-gomez · elon-musk · steve-jobs (archetype, deceased 2011)

Model Architects (7): andrej-karpathy · jared-kaplan · noam-shazeer · jason-wei · sebastian-raschka · [patrick-lewis](personas/patrick-lewis.md) · [thomas-wolf](personas/thomas-wolf.md)

Reasoning, RL, Agents (7): john-schulman · [noam-brown](personas/noam-brown.md) · hyung-won-chung · nathan-lambert · barret-zoph · karina-nguyen · sasha-rush

Alignment, Interp, Safety (9): chris-olah · paul-christiano · jan-leike · dan-hendrycks · stuart-russell · neel-nanda · lilian-weng · [beth-barnes](personas/beth-barnes.md) · [helen-toner](personas/helen-toner.md)

Theory and Science (9): yann-lecun · yoshua-bengio · geoffrey-hinton · [john-jumper](personas/john-jumper.md) · percy-liang · christopher-manning · [yejin-choi](personas/yejin-choi.md) · sara-hooker · aleksander-madry

Multimodal, Embodied (7): fei-fei-li · pieter-abbeel · sergey-levine · chelsea-finn · robin-rombach · aditya-ramesh · prafulla-dhariwal

Systems, Kernels, Serving (7): tri-dao · bryan-catanzaro · [andrew-feldman](personas/andrew-feldman.md) · albert-gu · horace-he · woosuk-kwon · tim-dettmers

Files in this team

superintelligenceTeam/
├── SKILL.md This file — user-facing entry.
├── registry.json Machine source of truth. Read by slash commands.
├── EXPANSION.md (Phase 2) Gap-analysis and roster-expansion candidates.
├── templates/
│ ├── persona.md Schema source-of-truth for every persona file.
│ └── convene.md Multi-persona session template.
├── personas/ 59 *.md files, one per persona. YAML frontmatter + 6 narrative sections.
├── cells/ 8 *.md cell summaries.
├── research/ 59 directories, one per persona. Raw research dumps so future re-syntheses don't recrawl.
└── scripts/
    └── build_registry.py Regenerates registry.json from persona frontmatter. Run after any persona edit.

<a id="slash-commands"></a>

Slash commands (Phase 3 — installed)

22 slash command files live at ~/.claude/commands/superintelligenceTeam*.md (user-global install, parallel to the existing /team family). Each is auto-registered as a discoverable Skill — no separate SKILL.md registration is needed. The architecture is orchestrator-first: every action verb invokes the orchestrator to pick a custom 16-32 persona team and gate on user approval before executing.

Dispatcher (1)

Orchestrator (1)

Identity surface — explicit overrides (4)

Roster management (1)

Action surface (15) — orchestrator-first by default

All 15 invoke /superintelligenceTeam:orchestrate first unless --no-orchestrate, --cells, or --personas flag is supplied.

Global flags

Every command accepts these:

Conventions enforced by every command