model-card

Name: model-card
Author: Aperivue

208stars53forks

How to install model-card?

aperivue/medsci-skills/model-card

$npx -y skills add Aperivue/medsci-skills --skill model-card

Installs into the current project.

Summary

Generate the documentation an engineer-built medical-imaging model must carry — a Model Card (Mitchell et al. 2019), a Datasheet for its dataset (Gebru et al. 2021), and a METRIC-informed data-quality pass — filled from user-supplied facts, then verify every required section is p

Files · 1

View on GitHub

SKILL.md

1	# Model-Card Skill
2
3	## Purpose
4
5	This skill produces the documentation an engineer-built medical-imaging model must carry: a
6	Model Card (intended use, out-of-scope use, training data, per-subgroup performance, caveats), a
7	Datasheet for its dataset (provenance, composition, collection, labelling, consent), and a
8	METRIC-informed data-quality pass. It fills the templates from facts the user supplies — it
9	never invents a number, a provenance detail, a consent status, or a licence — and ships a deterministic
10	gate that no required section is missing or left as an unfilled `[NEEDS INPUT]` placeholder.
11
12	It is the reporting seam of the model-engineering lane: after `/model-validation` audits the design
13	and `/model-evaluation` produces the numbers, this skill records them in a portable, auditable card that
14	`/write-paper` and `/check-reporting` consume. It mirrors `/version-dataset` structurally (generate +
15	deterministic verify).
16
17	## When to use
18	- A trained model needs a Model Card / Datasheet for a repo, Hugging Face card, or manuscript supplement.
19
20	## When NOT to use
21	- Auditing the validation design / metrics → `/model-validation`, `/model-evaluation`.
22	- Versioning the dataset bytes → `/version-dataset`; tabular variable docs → `/generate-codebook`.
23	- Item-by-item reporting-guideline compliance of the manuscript → `/check-reporting`.
24	- Building / training the model → `/model-scaffold`.
25
26	## Workflow
27
28	### Phase 1 — Collect the facts
29	Gather, from the user / the model's developers: task + architecture + provenance + licence; intended use
30	and out-of-scope use; training and evaluation cohorts; the reference standard and inter-reader agreement;
31	overall and per-subgroup performance; data collection, consent, and de-identification. Anything not
32	supplied stays `[NEEDS INPUT]` — never guess.
33
34	### Phase 2 — Fill the Model Card
35	Copy `${CLAUDE_SKILL_DIR}/references/model_card_template.md` to `MODEL_CARD.md` and fill each section
36	from the facts. Keep the headings. Numbers come only from `/model-evaluation` / executed results.
37
38	### Phase 3 — Fill the Datasheet
39	Copy `${CLAUDE_SKILL_DIR}/references/datasheet_template.md` to `DATASHEET.md` and fill the seven
40	question groups (Motivation, Composition, Collection, Preprocessing/Labeling, Uses, Distribution,
41	Maintenance).
42
43	### Phase 4 — METRIC data-quality pass
44	Walk `${CLAUDE_SKILL_DIR}/references/metric_dimensions.md` (completeness, correctness, consistency,
45	representativeness, timeliness, provenance, label provenance, fairness/coverage, leakage safety) and
46	record each finding in the Datasheet. Anything that affects the headline metric's validity is also a
47	`/model-validation` finding — cross-check there.
48
49	### Phase 5 — Verify completeness (deterministic gate)
50	```bash
51	python3 ${CLAUDE_SKILL_DIR}/scripts/check_model_card_complete.py \
52	--card MODEL_CARD.md --datasheet DATASHEET.md --strict
53	```
54	`MISSING_SECTION` / `EMPTY_REQUIRED_SECTION` must be zero before the card ships.
55
56	### Phase 6 — Hand off
57	Carry the card into `/write-paper` (the Methods / supplement reference it), `/check-reporting`
58	(CLAIM 2024 / TRIPOD+AI item audit of the manuscript), and `/self-review`.
59
60	## Anti-Hallucination
61
62	- Never invent evaluation numbers, subgroup results, or dataset provenance. Every figure comes from
63	`/model-evaluation` or the user's executed results; every provenance / consent / licence statement is
64	user-confirmed. Unknown → `[NEEDS INPUT]`, which the gate flags.
65	- Never mark a section complete without user-supplied content, and never auto-fill a placeholder to
66	pass the gate.
67	- Never assert a licence or consent status the user did not confirm.
68	- The gate checks presence, not truth — a complete card can still contain a wrong number; validity
69	is `/model-validation` and the human's responsibility.
70
71	## Deterministic gate
72	`scripts/check_model_card_complete.py` — verifies every required Model Card / Datasheet section is
73	present and non-empty (stdlib, network-free). Reproducible challenge:
74	`bash ${C

People who liked this also viewed

6 picks

Type

model-card

208stars53forks

View on GitHub

How to install model-card?

aperivue/medsci-skills/model-card

$npx -y skills add Aperivue/medsci-skills --skill model-card

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Model-Card Skill
2
3	## Purpose
4
5	This skill produces the documentation an engineer-built medical-imaging model must carry: a
6	Model Card (intended use, out-of-scope use, training data, per-subgroup performance, caveats), a
7	Datasheet for its dataset (provenance, composition, collection, labelling, consent), and a
8	METRIC-informed data-quality pass. It fills the templates from facts the user supplies — it
9	never invents a number, a provenance detail, a consent status, or a licence — and ships a deterministic
10	gate that no required section is missing or left as an unfilled `[NEEDS INPUT]` placeholder.
11
12	It is the reporting seam of the model-engineering lane: after `/model-validation` audits the design
13	and `/model-evaluation` produces the numbers, this skill records them in a portable, auditable card that
14	`/write-paper` and `/check-reporting` consume. It mirrors `/version-dataset` structurally (generate +
15	deterministic verify).
16
17	## When to use
18	- A trained model needs a Model Card / Datasheet for a repo, Hugging Face card, or manuscript supplement.
19
20	## When NOT to use
21	- Auditing the validation design / metrics → `/model-validation`, `/model-evaluation`.
22	- Versioning the dataset bytes → `/version-dataset`; tabular variable docs → `/generate-codebook`.
23	- Item-by-item reporting-guideline compliance of the manuscript → `/check-reporting`.
24	- Building / training the model → `/model-scaffold`.
25
26	## Workflow
27
28	### Phase 1 — Collect the facts
29	Gather, from the user / the model's developers: task + architecture + provenance + licence; intended use
30	and out-of-scope use; training and evaluation cohorts; the reference standard and inter-reader agreement;
31	overall and per-subgroup performance; data collection, consent, and de-identification. Anything not
32	supplied stays `[NEEDS INPUT]` — never guess.
33
34	### Phase 2 — Fill the Model Card
35	Copy `${CLAUDE_SKILL_DIR}/references/model_card_template.md` to `MODEL_CARD.md` and fill each section
36	from the facts. Keep the headings. Numbers come only from `/model-evaluation` / executed results.
37
38	### Phase 3 — Fill the Datasheet
39	Copy `${CLAUDE_SKILL_DIR}/references/datasheet_template.md` to `DATASHEET.md` and fill the seven
40	question groups (Motivation, Composition, Collection, Preprocessing/Labeling, Uses, Distribution,
41	Maintenance).
42
43	### Phase 4 — METRIC data-quality pass
44	Walk `${CLAUDE_SKILL_DIR}/references/metric_dimensions.md` (completeness, correctness, consistency,
45	representativeness, timeliness, provenance, label provenance, fairness/coverage, leakage safety) and
46	record each finding in the Datasheet. Anything that affects the headline metric's validity is also a
47	`/model-validation` finding — cross-check there.
48
49	### Phase 5 — Verify completeness (deterministic gate)
50	```bash
51	python3 ${CLAUDE_SKILL_DIR}/scripts/check_model_card_complete.py \
52	--card MODEL_CARD.md --datasheet DATASHEET.md --strict
53	```
54	`MISSING_SECTION` / `EMPTY_REQUIRED_SECTION` must be zero before the card ships.
55
56	### Phase 6 — Hand off
57	Carry the card into `/write-paper` (the Methods / supplement reference it), `/check-reporting`
58	(CLAIM 2024 / TRIPOD+AI item audit of the manuscript), and `/self-review`.
59
60	## Anti-Hallucination
61
62	- Never invent evaluation numbers, subgroup results, or dataset provenance. Every figure comes from
63	`/model-evaluation` or the user's executed results; every provenance / consent / licence statement is
64	user-confirmed. Unknown → `[NEEDS INPUT]`, which the gate flags.
65	- Never mark a section complete without user-supplied content, and never auto-fill a placeholder to
66	pass the gate.
67	- Never assert a licence or consent status the user did not confirm.
68	- The gate checks presence, not truth — a complete card can still contain a wrong number; validity
69	is `/model-validation` and the human's responsibility.
70
71	## Deterministic gate
72	`scripts/check_model_card_complete.py` — verifies every required Model Card / Datasheet section is
73	present and non-empty (stdlib, network-free). Reproducible challenge:
74	`bash ${C

People who liked this also viewed

6 picks

Type