- Home
- Skills
- Data & Databases
- CRISP-DM Phase 1 — Business Understanding
Works with the AI Agents you already use
CRISP-DM Phase 1 — Business Understanding
Earn the right to build the pipeline before touching data.
Free
CRISP-DM Phase 1 — Business Understanding
Example session with this skill installed
Frame this data project: we want to use satellite imagery and GIS to map urban heat islands in Barcelona.
- Read your context and instructions
- Compiled the crisp-dm phase 1
Phase 1 — Business Understanding
Refused to proceed as stated. "Use satellite imagery and GIS" is a technology, not a decision. Two blocking unknowns:
- Who decides what differently once this exists?
- What changes on the ground if the map is wrong?
Rewriting the subject as a decision, then producing the twelve artifacts under phase-1/:
| Artifact | Content |
|---|---|
decision-statement.md | Named role at a named organization; the decision, its cadence, the mechanism it feeds. Not "stakeholders". |
heilmeier.md | The DARPA catechism, answered — including "who cares?" and "what are the mid-term and final exams?" |
five-whys.md | Chain from symptom to root cause. |
decision-unit.md | Spatial, temporal and thematic unit stated as numbers (30 m pixel, annual, census tract) — not adjectives. |
output-spec.md | Schema, units, CRS (EPSG:25831), license, refresh cadence. |
success-criteria.md | Numbers with deadline and owner. Adjectives refused — "actionable", "useful", "better" blocked on entry. |
situation.md | What exists, what does not, who has it. |
risks.md | Register with a cancellation criterion: at least one row marked "Triggers cancellation = Yes". Its absence is itself a defect. |
glossary.md | Every contested term defined once. |
product-card-draft.md | Includes at least two named out-of-scope uses. |
project-plan.md | Phases, artifacts, owners, dates. |
exit-checklist.md | Signed. An unticked box means Phase 1 is not done. |
Gate run
$ python scripts/gate_check.py --phase 1 --root .
[FAIL] phase-1/success-criteria.md contains adjectives: "actionable"
[FAIL] phase-1/risks.md has no row with "Triggers cancellation = Yes"
[FAIL] phase-1/decision-statement.md: decision-maker is "stakeholders"
GATE 1: FAIL (3 blocking defects) — exit code 1
After fixing
$ python scripts/gate_check.py --phase 1 --root .
GATE 1: PASS — 12/12 artifacts present and non-empty
Next: Phase 2 (data understanding) — not included in this edition.
Outcome
Phase 1 holds. The project has earned the right to go and get data — and can state, in writing, what would make it stop.
Connects securely to your tools. The creator never sees your data.
About this skill
Most data projects fail before the first download — not technically, definitionally. The team builds exactly what was asked for, then discovers at evaluation that what was asked for was not what was needed. This skill blocks that failure at the only point where it is cheap to fix: the start.
What it does. Forces twelve specific artifacts onto disk, each with a binary exit check: a decision statement with a named decision-maker (not "stakeholders"), a decision unit stated as numbers, an output specification (schema, units, CRS, license), numerical success criteria — numbers, deadlines, owners; adjectives are refused — a situation inventory, the Heilmeier answers, a Five Whys chain, a risk register with a cancellation criterion, a glossary, a product card with named out-of-scope uses, a project plan, and a signed exit checklist.
What it refuses. Solutionism ("we want to use LLMs/GIS to…" — that is a tool, not a problem). "Stakeholders" as a decision-maker. Adjective success criteria ("useful", "actionable"). Pre-emptive data hunting. Projects with no written reason to be killed.
Executable gates. Ships with two Python scripts (standard library only): crisp_status.py tells you where the project stands; gate_check.py refuses advancement with a non-zero exit code until every artifact exists. Binary, not vibes.
Built on. CRISP-DM 1.0 (Chapman et al. 2000), CRISP-ML(Q) quality assurance (Studer et al. 2021), the Heilmeier catechism (DARPA), SMART criteria, project pre-mortems (Klein). Full citations included.
Works with any coding agent. Claude Code, Codex CLI, Hermes, Cursor, Gemini CLI, Copilot — anything that reads the Agent Skills convention. Single-agent friendly: per-step "specialist lenses" instead of sub-agent dependencies.
Free component of crisp-for-data — the complete Phase 0–6 suite (data understanding, preparation with a YAML data contract, the analytical core, evaluation with a go/no-go memo, FAIR/DOI release handoff) by the same author. Phase 1 here is the gate all of it runs on.
Covers CRISP-DM Phase 1 only. Refuses data exploration or modeling. Requires project-root write access for artifact generation. Validation scripts require Python.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires