prompt-hub

Prompt Hub Skill

When this skill is loaded, you have access to the FrankX Prompt Hub: a 13-agent team that handles every prompt-engineering operation across Claude, GPT, Gemini, and OSS models, plus an IFS / psychometric introspection layer.

How to invoke

For any prompt-engineering ask, dispatch @prompt-conductor via the Task tool with the user's verbatim request. The Conductor maps it to one of 8 canonical flows and composes 2-5 specialists.

Do not author prompts yourself when this skill is loaded. Route through the Hub.

The 13 agents

Conductor (1)

• @prompt-conductor — Opus composer; routes all asks.

Lab specialists (4) — encapsulate per-lab quirks

• @prompt-claude-specialist — XML tags, prefill, extended thinking
• @prompt-gpt-specialist — developer role, Structured Outputs, contradiction audit
• @prompt-gemini-specialist — system-at-top, native grounding, "think very hard"
• @prompt-oss-specialist — Llama / Mistral / Qwen / R1 chat templates

Core builders (3)

• @prompt-architect — designs new prompts from blank
• @prompt-optimizer — refines existing prompts (wraps /po)
• @prompt-evaluator — wraps promptfoo (MIT), scores prompts

Library curators (2)

• @prompt-librarian — owns prompt-library repo; ranks, tags, attributes
• @prompt-harvester — bulk-imports from Fabric / awesome-* repos

Safety + psyche (3)

• @prompt-red-team — adversarial probes; load-bearing publish gate
• @prompt-psyche-cartographer — IFS introspection, voice modes, maps not unburdens
• @prompt-psychometrist — IPIP-50, PVQ, ECR-R, VIA, Enneagram (lenses not verdicts)

The 8 flows

Flow	Trigger	Sequence
flow-design	"design a prompt for X"	architect → lab-spec → red-team → evaluator
flow-optimize	"optimize this prompt" / /po	optimizer → lab-spec → evaluator
flow-evaluate	"evaluate / test / score this prompt"	evaluator → red-team
flow-harvest	"import from Fabric / harvest awesome-*"	harvester → red-team → librarian
flow-curate	"rerank / rebuild library"	librarian → optimizer → evaluator
flow-introspect	"IFS session / part of me / journal"	cartographer (solo)
flow-profile	"Big Five / values map / attachment"	psychometrist → cartographer
flow-knowledge-base	"RAG prompts / ingestion set"	architect → librarian → evaluator

Invariants (load-bearing)

1. Red Team gates every publish. No pattern lands in prompt-library without red_team.status: pass.
2. Evaluator scores every publish. No pattern lands with eval.score < 3.5.
3. Voice gate runs on every output. lib/voice/frankx-voice.ts checks banned phrases.
4. Crisis triggers abort psyche flows. Routes to 988 / Samaritans / Befrienders.
5. No clinical content in Cartographer. Mapping only, never unburdening.
6. Attribution mandatory on imports. Provenance frontmatter + ATTRIBUTION.md entry.

Where things live

• Master spec: docs/superpowers/specs/2026-05-13-prompt-hub-design.md
• Agent files: .claude/agents/prompt-*.md
• This skill: .claude/skills/prompt-hub/SKILL.md
• Command: .claude/commands/prompt-hub.md
• Schema + types: lib/prompt-hub/types.ts
• Voice gate: lib/voice/frankx-voice.ts
• Engine repo (pre-extract): repos/prompt-engine/
• Library repo (pre-extract): repos/prompt-library/
• Public pages: app/prompts/ + app/prompt-library/

Composition into existing systems

• /po = alias for flow-optimize
• /superintelligence = called BY conductor for cross-lab master-prompt synthesis
• /acos-meta = should document this Hub as substrate slot
• Second Brain OS = consumed by Cartographer + Psychometrist for cross-session recall (never duplicated)
• Pillars 1-6 = all consume the Hub for their per-pillar prompt needs

Anti-patterns

• Authoring prompts inline when this skill is loaded — route through Conductor.
• Skipping Red Team because "this looks safe" — gate is non-negotiable.
• Mixing introspection + library flows — strict separation.
• Inventing new lab specialists when an existing one covers the lab — extend, don't sprawl.
• Lifting patterns from closed-source marketplaces (PromptHub / PromptBase) — never.

Reference

• Pattern schema: repos/prompt-engine/schema/pattern.schema.json
• promptfoo: https://promptfoo.dev (MIT)
• Fabric (inspiration): https://github.com/danielmiessler/fabric (MIT)
• Anthropic prompt eng: https://platform.claude.com/docs/en/docs/build-with-claude/prompt-engineering/overview
• OpenAI GPT-5 guide: https://developers.openai.com/cookbook/examples/gpt-5/gpt-5_prompting_guide
• Gemini strategies: https://ai.google.dev/gemini-api/docs/prompting-strategies
• IFS canon: Schwartz, No Bad Parts
• Anthropic introspection paper: https://www.anthropic.com/research/introspection