Horizontal substrate · 13 agents · 2 OSS repos · 8 flows

The Prompt Hub

Every prompt evaluated, attributed, lab-tagged, red-teamed. A 13-agent team handling design, optimization, evaluation, harvest, curation, introspection, and psychometric profiling across Claude, GPT, Gemini, and open-source models.

Built on the same Personal AI CoE framework as ACOS. Open source. No vibes. No clinical drift.

Three commitments

Every published prompt walks the same gate

Designed by the architect, shaped by the lab specialist, optimized for token economy, red-teamed against jailbreaks and prompt injection, evaluated against declarative test cases via promptfoo. Score, verdict, and probe count live in each pattern's frontmatter. No vibes.

Every contributor uses the same schema

Frontmatter is the contract: id (verb_topic), version (semver), lane (Claude / GPT / Gemini / OSS / cross-lab), category (analyze / create / extract / ...), provenance (source, URL, attribution, license), eval score, red-team status. Fabric-style folder-per-pattern, promptfoo evals colocated. Bulk imports from awesome-chatgpt-prompts (CC0) and awesome-claude-prompts (MIT) follow the same rules — full attribution, license verified, quality-gated.

Every user can map their own patterns

The Psyche Cartographer offers IFS-based introspection with Socratic / Stoic / Jungian / IFS-Self voice modes. The Psychometrist administers IPIP-50, Schwartz PVQ, ECR-R, VIA, Enneagram. Both frame results as lenses, never verdicts. Hard boundaries: maps only, never unburdens. Crisis triggers route to 988 / Samaritans / Befrienders. Inspired by Mindsera and Anthropic's introspection research.

The 13-agent team

One top-level Opus composer routes every ask to 2-5 specialists. Lab quirks (Claude XML, GPT Structured Outputs, Gemini grounding, OSS chat templates) each get their own home.

Composer

@prompt-conductorOpus composer; routes every ask to a flow

Lab specialists

@prompt-claude-specialistXML tags, prefill, extended thinking
@prompt-gpt-specialistdeveloper role, Structured Outputs, contradiction audit
@prompt-gemini-specialistsystem-at-top, native grounding
@prompt-oss-specialistLlama / Mistral / Qwen / R1 chat templates

Core builders

@prompt-architectdesigns new prompts from blank
@prompt-optimizerrefines existing prompts
@prompt-evaluatorwraps promptfoo; scores prompts

Library curators

@prompt-librarianowns the corpus; ranks, tags, attributes
@prompt-harvesterbulk-imports from Fabric / awesome-* repos

Safety + psyche

@prompt-red-teamadversarial probes; publish gate
@prompt-psyche-cartographerIFS introspection; maps not unburdens
@prompt-psychometristIPIP-50 / PVQ / ECR-R / VIA — lenses not verdicts

The 8 canonical flows

Flow	Trigger	Sequence
flow-design	design a prompt for X	architect → lab-specialist → red-team → evaluator
flow-optimize	optimize this prompt / /po	optimizer → lab-specialist → evaluator
flow-evaluate	evaluate / 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 / journal	cartographer (solo)
flow-profile	Big Five / values map	psychometrist → cartographer
flow-knowledge-base	RAG prompts / ingestion	architect → librarian → evaluator

Open source

frankxai/prompt-engine

MIT · scaffolded

The 13-agent team + tooling. Bootable Node project.

View repo →

frankxai/prompt-library

MIT + CC0 imports with attribution · scaffolded

The Library-of-Alexandria corpus. Curated, ranked, attributed.