Computational creativity and open-ended problem solving for the Howl ecosystem.
HowlCreate exists to answer:
What could we do?
It deliberately explores possibilities before another system decides how to execute them.
HowlCreate is a computational creativity engine designed for divergent and convergent creative reasoning. Rather than issuing single-pass LLM brainstorming prompts, HowlCreate executes a disciplined, multi-operator search: extracting hidden assumptions, inverting constraints, analogizing across distant domains, branching independently to avoid premature convergence, challenging proposals through adversarial critique, synthesizing disparate ideas, and converging toward a defensible, lineage-tracked set of finalists.
A strict architectural boundary separates HowlCreate from the rest of the Howl ecosystem:
| System | Role | Core Question |
|---|---|---|
| HowlCreate | Imagination, reframing, lateral exploration, synthesis, concept evolution | What could we do? |
| HowlFrame | Structured reasoning, evidence verification, claims, uncertainty | What do we know, why do we believe it, and what remains uncertain? |
| HowlPlane | Planning, orchestration, task decomposition, agent routing, execution graphs | How do we get this done? |
| HowlChangeOps | Authority boundary, release gating, safety approvals, rollback | Are we allowed to do this, and can we do it safely? |
| HowlWriter | Source-grounded writing, humanization, voice adaptation, communication | How should we communicate this? |
Architectural Guardrail: HowlCreate does not plan execution or orchestrate agent tasks. It generates and evaluates creative possibilities. Once a concept is selected, it can be handed off to HowlFrame (for verification of claims and evidence needs) and HowlPlane (for execution planning).
Creative reasoning follows a disciplined divergent-then-convergent pipeline:
PROBLEM
│
▼
UNDERSTAND & EXTRACT ASSUMPTIONS
│
▼
REFRAME (Multiple Lenses & Perspectives)
│
▼
DIVERGE (Independent Branching)
│
▼
MUTATE (Constraints, Analogies, Extremes, Substitutions)
│
▼
CROSS-POLLINATE & SYNTHESIZE (Forced Combinations)
│
▼
CHALLENGE (Adversarial Critique & Brittleness Analysis)
│
▼
EVALUATE (Multi-Dimensional Scoring & Uncertainty Tracking)
│
▼
CONVERGE (Preserving Distinct Strategies & Wildcards)
│
▼
OUTPUT CONCEPTS (Lineage-Tracked Finalists)
- Divergence Before Convergence: Never rank ideas immediately after generation. Premature evaluation suffocates exploration. The search budget is first spent discovering distant branches.
- Epistemic Truthfulness: Facts, assumptions, speculation, hypotheses, analogies, and predictions are explicitly labeled. Speculation is never masqueraded as factual evidence.
- Lineage Preservation: Every concept is a node in a directed acyclic graph (DAG) recording its origin, parent ideas, creative operator, mutated assumptions, and survival rationale.
- Provider-Agnostic Search: Built-in support for deterministic/mock execution (for offline testing and CI), local Ollama models, and OpenAI-compatible endpoints.
# Clone the repository
git clone https://github.com/howlcipher/howlcreate.git
cd howlcreate
# Install in development mode
pip install -e ".[dev]"# Run a full creative exploration pipeline
howlcreate explore "How could the Howl ecosystem create meaningful remote-work opportunities?"
# Extract and invert hidden assumptions
howlcreate assumptions "How to build an offline-first distributed database"
# Generate multi-perspective reframings
howlcreate reframe "Developer onboarding takes three weeks"
# Inspect a past run and lineage graph
howlcreate show <run-id>
howlcreate graph <run-id>
# Export structured JSON for downstream systems (HowlPlane / HowlFrame)
howlcreate export <run-id> --output concept.json# Run test suite
pytest -v
# Static checks
flake8 src/ tests/ --select=F821,F401,E9
# Verification build
python -m compileall src/ tests/MIT License. See LICENSE for details.