jack-h-park

PM Intelligence Engine — Data skill for Claude Code

Data community

This platform automates a 7-stage PM decision workflow.

How to install PM Intelligence Engine

This entry records only its repository, not the path inside it, so there is no exact command to give. Open jack-h-park/pm-intelligence-engine and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What PM Intelligence Engine does

This platform automates a 7-stage PM decision workflow. Each stage is a discrete step that transforms the input it receives and writes its output to the database before passing control to the next stage.

Alternatives in Data

HTML Anything — ✨ The agentic HTML editor — your local AI agent writes the HTML, you ship it 8.5k ★
Clinical Decision Support — Evidence-based clinical decision support tools 6.4k ★
Check Data Quality — Renamed → /portaljs-check-data-quality 2.3k ★

README

PM Intelligence Engine

A provider-neutral Python service that turns incoming signals into a traceable, human-gated decision workflow. It is designed to work with operator-supplied context and integrations; no private context or credentials are included in this repository.

Workflow

`SENSE → DECIDE → LEARN`

SENSE normalizes an incoming signal.
DECIDE runs seven typed stages, including four independent evaluations and explicit human approval gates.
LEARN persists the run and exports structured artifacts for downstream systems.

The engine supports Claude and OpenAI through the `LLMProvider` protocol, SQLite persistence, a FastAPI API, and a regression/evaluation harness.

Quick start

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
# Set LLM_PROVIDER and the matching provider API key in .env.
uvicorn app.api.main:app --reload

All authenticated endpoints require `PM_PLATFORM_API_TOKEN`; `/health` is the only unauthenticated endpoint. The default context and wiki roots are local directories. Provide your own context files or override the paths in `.env`.

Development

pytest -q
ruff check app tests eval
python eval/runner.py

The evaluation fixtures are examples only. Replace them with domain-appropriate fixtures before using the engine for production decisions. Do not commit API keys, generated databases, run archives, or private context repositories.

Repository layout

app/       FastAPI routes, stages, models, storage, and LLM adapters
eval/      Regression scenarios and quality rubrics
tests/     Unit and integration tests
docs/      API, architecture, and design contracts