Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentscrisp-for-data — Complete CRISP-DM (Phases 0–6)
The full CRISP-DM cycle as one skill, gates included.
$49
crisp-for-data — Complete CRISP-DM (Phases 0–6)
Example session with this skill installed
Run Phase 5 on this pipeline. We built a composite tree-canopy vulnerability index for Barcelona's 10 districts, weighted 40% canopy loss, 35% surface temperature, 25% imperviousness. The result looks right. Can we ship it?
- Read your context and instructions
- Compiled the crisp-for-data
Phase 5 — Evaluation. Default verdict: NEEDS WORK.
Evaluation here is not "are the model metrics good". It is "does this answer the decision the project committed to in Phase 1". So the first artifact read is not your output — it is phase-1/success-criteria.md.
Check 1 — Business fit against the project's own criteria
Committed in Phase 1: "Rank the 10 districts for the 2027 planting budget; the Deputy Director must be able to defend the ordering in a council session."
Delivered: a ranking.
MATCH. But the Phase-1 criterion also required "stable under ±10% weight perturbation". Not tested.
FAIL.
Check 2 — Sensitivity analysis (mandatory on every weight)
Re-ran the index across ±10% on all three weights (100 draws). District ordering is not stable:
| District | Base rank | Perturbed rank | Verdict |
|---|---|---|---|
| Eixample | 1 | 1 | stable |
| Sant Marti | 2 | 4 | SWINGS |
| Ciutat Vella | 3 | 2 | SWINGS |
| Sant Andreu | 8 | 6 | SWINGS |
3 of 10 adjacent pairs flip. The headline claim — "we ranked them" — does not survive its own error bars. The honest headline is: "we identified a top group of 3 and a bottom group of 4; the middle 3 are not separable with this data."
Check 3 — Leakage and validity
Surface temperature is derived from the same Landsat scene used to build the canopy-loss layer. Shared acquisition, shared atmospheric correction. Declared? No. Not fatal — but it must be written into the limitations register, not left for a reviewer to find.
FAIL (undeclared).
Check 4 — The Monday test
"What does the Deputy Director DO on Monday that they could not do on Friday?" — requires a named human's verbatim answer. A summary of your own reasoning is not accepted.
NOT SUPPLIED. FAIL.
Check 5 — Pre-mortem
"It is 2028 and the planting budget was spent on the wrong districts. Write the post-mortem." Three credible failure paths produced; two are not currently mitigated.
Verdict: ITERATE (not GO, not KILL)
- Blocking: sensitivity instability is a real claim defect; the Monday test has no named human behind it.
- Non-blocking but required: declare the shared-acquisition issue in the limitations register.
Written to phase-5/
go-no-go-memo.md— verdict: ITERATE (machine-readable block)limitations-register.md— incl. shared-acquisition dependencyintended-use-statement.md— who may use this, for what, and what it must not be used forconditions-for-non-use.md— explicit stop conditionsstakeholder-walkthrough.md— the script to present this honestly
A clean no-go is a valid outcome. Shipping requires the named decision-maker's own answer to the Monday test — not the analyst's.
Connects securely to your tools. The creator never sees your data.
About this skill
CRISP-DM 1.0, translated for the projects that dominate real work — pipelines whose deliverable is a data product (a curated dataset, a scoring map, a network analysis, a report), not a deployed ML model and not a UI.
Research DNA. Distilled from a 30,000-word academic synthesis with 100+ verified citations — CRISP-DM 1.0 (Chapman 2000), CRISP-ML(Q) (2021), Gebru datasheets, Mitchell model cards, OECD/JRC composite-indicator methodology, Saltelli sensitivity analysis, Kapoor & Narayanan on leakage, Croissant metadata, FAIR. The bibliography ships with the skill.
Gates that refuse to open. Each phase refuses to start without the previous phase's artifacts, and the exit criteria are binary checklists enforced by scripts (gate_check.py, verify_contract.py) — not prose. The default evaluation verdict is NEEDS WORK; shipping requires a named decision-maker's verbatim answer to the "Monday test". A clean no-go is a valid outcome.
Phases people skip, done properly. Phase 0 (data acquisition: formal requests, license terms, ethics screening, fallbacks), a close-out companion for Phase 2 (SHA-256 ingestion logs, executable schemas, MAUP declarations, Croissant JSON-LD sidecars), a YAML data contract between preparation and modeling, and Phase 6 reframed as release handoff (lockfiles, FAIR, separate code/data/doc licenses, DOI minting, monitoring and ownership plans).
Phase 4 is the analytical core, not an ML tutorial. A decision tree routes your question to the right family — ranking to composite indicator; pattern to LISA/Gi*; structure to graph analysis; counterfactual to simulation; prediction to ML only through an explicit gate — then enforces pre-registered test design and mandatory sensitivity analysis.
Agent-native and agent-agnostic. Trigger phrases for automatic invocation; phases load on demand for a small context footprint; works on Claude Code, Codex CLI, Hermes, or any Agent Skills reader.
A fully worked example. Filled decision-log tables (selection, cleaning, construction, integration, format) and a complete data contract, modeled on a real Barcelona urban-ecology pipeline.
In the box: one skill — router + 8 phase files + 14 consolidated deep-dive files + a quality-gates overlay + a worked example + 3 executable gate scripts. Install by unzipping into your agent's skills directory.
Who it's for: data analysts and consultants shipping defensible data products; urban planning, environmental, AEC and public-sector analytics; research teams that must show exactly how every number was produced.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scanned
Verified clean 3 days ago
- Passed all security checks, Safe to install
Requires
Python
Git