Structured multi-criteria decisions, confidence, and explainable tradeoffs.

ADI Decision Engine is an OpenClaw skill bundle for structured multi-criteria decision analysis, turning messy tradeoff problems into ranked, auditable recommendations.

Built for vendor selection, route planning, hiring shortlists, software/tool comparison, procurement, and research method evaluation.

What It Is

This repository packages adi-decision as a professional OpenClaw skill. The skill accepts either:

• a structured ADI-style decision request with options, criteria, constraints, and evidence
• a freeform decision brief that first needs to be normalized into a request skeleton

The skill then validates the request, runs deterministic ADI locally, and returns:

• ranked options
• overall confidence
• explanation of the winning tradeoff
• constraint effects
• sensitivity and stability signals
• counterfactual guidance when available

Skill Snapshot

Structured input Convert vague tradeoff problems into explicit options, criteria, weights, and constraints.

Deterministic output Run a reproducible ADI scoring pipeline instead of improvising subjective rankings.

Explainable rationale Return confidence, constraint effects, contributions, and counterfactual guidance.

Why It Exists

Many decision-support prompts are underspecified, subjective, or hard to audit after the fact. ADI Decision Engine narrows that ambiguity by enforcing an explicit schema and deterministic scoring flow.

This repository focuses on a public, discoverable OpenClaw packaging layer rather than re-implementing the ADI core itself.

Key Capabilities

• structured decision support for tradeoff-heavy problems
• multi-criteria decision analysis with auditable weights and constraints
• deterministic ranking with explainable recommendations
• confidence-aware scoring and evidence handling
• policy-driven decisions with balanced, risk_averse, and exploratory
• request normalization for vague problem statements
• bundled examples across procurement, routing, hiring, research, and tooling

Example Decision Domains

Domain	Typical criteria	Why ADI fits
Vendor selection	cost, quality, lead time, contractual risk	Clear tradeoffs and defensible procurement rationale
Route planning	time, cost, walking burden, transfers	Explicit weighted tradeoffs instead of vague preference guessing
Hiring shortlists	skill fit, communication, delivery risk, compensation	Transparent scoring and stronger decision hygiene
Tool selection	reliability, setup effort, support burden, extensibility	Policy-driven comparisons with confidence-aware ranking
Research methods	accuracy, cost, complexity, interpretability	Explorable tradeoffs with auditable assumptions

Repository Layout

.
├── README.md
├── LICENSE
├── CITATION.cff
├── docs/
│   └── images/
└── adi-decision-engine/
    ├── SKILL.md
    ├── agents/openai.yaml
    ├── scripts/
    ├── references/
    └── examples/

The OpenClaw bundle is isolated inside adi-decision-engine/ so that GitHub presentation and skill publishing remain separate concerns.

Using The Skill

The publishable bundle lives in adi-decision-engine/. The primary skill contract is in adi-decision-engine/SKILL.md.

Runtime helper scripts are bundled in adi-decision-engine/scripts:

• run_adi.py: validate and execute a full decision request
• validate_request.py: confirm a request is schema-valid
• normalize_problem.py: convert a plain-language brief into a request skeleton

Recommended workflow:

1. Start with a plain-language brief or a partial JSON object.
2. Use normalize_problem.py to turn it into a request skeleton.
3. Use validate_request.py once criteria, directions, and values are complete.
4. Use run_adi.py to produce the final ranked decision output.

Runtime Requirements

• python3
• either an importable adi-decision package or the adi CLI on PATH
• no API keys
• no network access for core skill execution

If the ADI runtime dependency is missing, the skill fails honestly and asks for package installation instead of fabricating scores.

Validation Status

This bundle is structured to support:

• OpenClaw frontmatter parsing
• aligned agents/openai.yaml metadata
• schema-aware request validation
• deterministic local execution through the upstream ADI package
• bundled golden examples for smoke testing

Local smoke verification already completed:

• freeform normalization smoke test
• structured validation on bundled examples
• full execution on vendor selection, route planning, and tool selection examples
• Python compilation for bundled scripts

OpenClaw UI Metadata

The skill ships with:

• branded icons in adi-decision-engine/assets/
• OpenClaw brand_color
• clean display_name, short_description, and default_prompt
• implicit invocation enabled through agents/openai.yaml

Suggested GitHub Topics

• openclaw
• decision-intelligence
• mcda
• decision-support
• weighted-scoring
• tradeoff-analysis
• option-ranking
• explainable-ai
• agent-tooling
• procurement

Citation

This repository includes CITATION.cff so that citation metadata is available once the repository is published.

Suggested positioning:

• cite the repository when you use the OpenClaw skill packaging
• cite adi-decision separately when you need to reference the underlying decision engine implementation

Publish Status

This repository is now published on GitHub:

• GitHub: dimgouso/adi-decision-engine_skill_openclaw

The OpenClaw bundle has also been published to ClawHub as:

• adi-decision-engine@0.1.0

At the time of this update, ClawHub reports the skill as temporarily hidden while its security scan is pending. A local readiness and publication log is available in docs/publish_checklist.md.

License

Released under the MIT License. See LICENSE.