Skip to content
Skills Library
Data & analytics · Building & coding

decision-engine

/decision-engine

The ADOP "AI clone" Decision Engine and guardrails. Load this whenever generating or promoting a data pipeline to encode the organization's architecture standards, tool-routing preferences, Cedar authorization policies, and the mandatory invariants. It governs HOW agents pick tools and what they are never allowed to do — so every builder produces a consistent architecture instead of a different one per engineer.

Licence
MIT
Last repository change
05.10.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
2
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 05.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

The ADOP "AI clone" Decision Engine and guardrails. Load this whenever generating or promoting a data pipeline to encode the organization's architecture standards, tool-routing preferences, Cedar authorization policies, and the mandatory invariants. It governs HOW agents pick tools and what they are never allowed to do — so every builder produces a consistent architecture instead of a different one per engineer.

Triggers from the description

AI clone
Original author description

The ADOP "AI clone" Decision Engine and guardrails. Load this whenever generating or promoting a data pipeline to encode the organization's architecture standards, tool-routing preferences, Cedar authorization policies, and the mandatory invariants. It governs HOW agents pick tools and what they are never allowed to do — so every builder produces a consistent architecture instead of a different one per engineer.

How it works

  1. Configuration layers
  2. 5-step tool selection hierarchy
  3. Mandatory invariants (starter set)
  4. How to use during a build
Original SKILL.md headings · 75 lines, 672 words
SKILL.md preview
---
name: decision-engine
description: The ADOP "AI clone" Decision Engine and guardrails. Load this whenever generating or promoting a data pipeline to encode the organization's architecture standards, tool-routing preferences, Cedar authorization policies, and the mandatory invariants. It governs HOW agents pick tools and what they are never allowed to do — so every builder produces a consistent architecture instead of a different one per engineer.
---

# Decision Engine + Guardrails

The Decision Engine is an **AI-encoded version of your enterprise architect**: it embeds the
org's guidelines, tech standards, and design philosophy into the build so the model *fills in
the blueprint, it doesn't draw it*. Fill in `config/` below with your org's real values; the
defaults are sane starting points.

## Configuration layers

Paths are relative to the plugin root and mirror the ADOP repository, so the first three are
byte-identical to their sources and `tests/unit/test_plugin_translation.py` fails if one drifts.

| File | Role |
|------|------|
| `TOOL_ROUTING.md` | Intent-to-tool mapping — 8 intents, each with `not_when` disqualifiers |
| `tool-registry/servers.yaml` | The MCP server inventory — 13 servers with tier, category and fallback |
| `tool-registry/invariants.yaml` | The 11 mandatory rules (BLOCK/WARN), enforced regardless of phase |
| `config/standards.md` | A template for your organisation's own tech choices — edit this one |

Only `config/standards.md` is meant to be edited in place: it is the deliberately-empty slot for
your architecture preferences. The other three are vendored copies, so change them upstream in
the repository and re-vendor. Editing a copy here would be silently reverted by the next
re-vendoring and would fail the drift check in the meantime.

Earlier versions of this plugin kept hand-maintained versions of all four, and they had drifted:
`servers.yaml` listed 12 servers instead of 13, omitting `core` and leaving `zone-scoped-kms`
and `no-credentials-in-code` with no server to route to; `invariants.yaml` listed 10 rules
instead of 11, omitting `iam-simulate-first` (BLOCK) while `TOOL_ROUTING.md` still routed an
`iam-simulate` intent and `standards.md` still required `simulate_principal_policy` before
source access; and `TOOL_ROUTING.md` carried 5 intents of 8, so whole phases — Bronze ingestion,
Silver and Gold transforms, lineage — had no routing at all and fell back on model priors.

## 5-step tool selection hierarchy

```
User Intent → 1: Context Gate → 2: Server Health → 3: Intent Match → 4: Fallback → 5: Invariants
```

1. **Context Gate** — *Am I a sub-agent?* If YES → STOP: generate files only, no MCP/AWS/CLI.
   If NO (orchestrator) → proceed. Enforced by Cedar policy + the `sub-agent-no-mcp` invariant.
2. **Server Health** — classify tools REQUIRED / WARN / OPTIONAL. REQUIRED failure = block;
   WARN = CLI fallback + log; OPTIONAL = skip gracefully.
3. **Intent Match** — match the operation to an intent entry in `TOOL_ROUTING.md`; honor the
   `not_when` negative filter.
4. **Fallback** — MCP first, always. If unavailable and REQUIRED → block + escalate to human;
   else CLI fallback + logged warning.
5. **Invariant Enforcement** — the rules below apply to *every* operation no matter what tool
   was chosen.

## Mandatory invariants (starter set)

| ID | Rule | Severity |
|----|------|----------|
| `lineage-always` | `--enable-data-lineage: true` on every Glue ETL job | BLOCK |
| `no-credentials-in-code` | Credentials via Secrets Manager / Airflow Connections only | BLOCK |
| `bronze-immutable` | Bronze data never modified after ingestion | BLOCK |
| `quality-gates` | Silver ≥ 80%, Gold ≥ 95%; critical failures block promotion | BLOCK |
| `sub-agent-no-mcp` | Sub-agents generate artifacts only; no MCP/AWS execution | BLOCK |
| `verify-deployment` | Confirm tables queryable after deploy | BLOCK |
| `audit-after-deploy` | Confirm audit logging active after deploy | BLOCK |
| `zone-scoped-kms` | Separate CMK for Bronze, Silver, Gold per workload | BLOCK |
| `mcp-first` | Use MCP tools first; CLI only when unavailable/errored | BLOCK |
| `log-cli-fallback` | Print a warning on every CLI fallback | WARN |

## How to use during a build
- **Sub-agents**: obey the Context Gate — you are always in the "generate files only" branch.
- **Orchestrator**: before any infra action, walk steps 2–5 and confirm no BLOCK invariant is
  violated. If one is, stop and surface it to the human.
- Keep the org's real preferences in `config/` (e.g. "prefer Athena over Redshift for ad-hoc",
  "PII operations must use designated KMS keys", cost thresholds, failover order). The clone
  applies them consistently so the same pipeline behaves the same across environments.

See `config/` for editable starter files.
Files in this package (2)
  • SKILL.md 5 KB
  • config/standards.md 1 KB

Versions

The database has no other skills named decision-engine.

Database snapshot: 13.11.2026 · package version 7241b18fedaf (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

No similar skills in the database.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.