scgpt

Embed and annotate single-cell expression data with scGPT, a foundation model for single-cell biology. Use this skill when: (1) Producing cell embeddings from an AnnData for clustering/integration, (2) Zero-shot or fine-tuned cell-type annotation, (3) Gene-level representation for perturbation/GRN tasks. For probabilistic single-cell models (scVI etc.), use the scvi-tools library.

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, largely actionable reference: executable code for the core embedding task, a prerequisites table, and an excellent troubleshooting section. The main defect is the Remote compute section, which mixes a coherent submitJob example with duplicated references and a confusing sentence about job-result reads, hurting both conciseness and clarity.

Suggestions

Deduplicate the Remote compute section: keep a single reference to the remote-compute-ssh/remote-compute-modal skills and rewrite or delete the confusing sentence about a final .result() read reporting 'suppressed' vs 'committed' follow-ups.

Replace the 'environment=...' placeholder with the concrete step for extracting the environment name from compute_details output so the submitJob example is copy-paste ready.

Add a cheap pre-flight validation step — e.g., report the fraction of adata.var genes found in the vocab before submitting a remote job — since 'unmatched genes are dropped' is the main silent failure mode.

Dimension	Reasoning	Score
Conciseness	Sections are lean and high-value (prerequisites table, terse gotchas, troubleshooting table), but the Remote compute section contains garbled, partly duplicated text — the confused sentence about '.result()' reporting 'suppressed' or 'committed', and the 'See the remote-compute-ssh ... skill' reference stated twice back-to-back.	4 / 5
Actionability	Concrete, near-executable snippets for vocab loading, embed_data, and job submission, plus a troubleshooting table with exact fixes (e.g., manifest key normalization 'name.lower().replace('-', '_')'). Minor gaps remain: the 'environment=...' placeholder and '/path/to/' stubs are not copy-paste ready.	4 / 5
Workflow Clarity	Sequences are clear (load vocab → embed → scanpy downstream; read compute_details → submitJob → retain job_id → poll .status()/.result()) with error-recovery guidance in Gotchas/Troubleshooting. However, no explicit pre-submit validation exists despite the warning that 'unmatched genes are dropped', leaving a minor checkpoint gap.	4 / 5
Progressive Disclosure	The body is well-sectioned with content appropriately placed (quick usage up front, gotchas and troubleshooting at the end) and clear pointers to the remote-compute-ssh / remote-compute-modal skills. The duplicated skill references and tangled orchestration paragraph are minor organization gaps.	4 / 5
	Total	16 / 20 Passed

Description

92%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: it states concrete capabilities, gives an explicit numbered 'Use when' trigger list, and cleanly disambiguates against scvi-tools. The only gap is a few natural synonyms (e.g., 'scRNA-seq', '.h5ad') that users could plausibly say.

Dimension	Reasoning	Score
Specificity	Multiple concrete actions are enumerated — 'Producing cell embeddings from an AnnData', 'Zero-shot or fine-tuned cell-type annotation', 'Gene-level representation for perturbation/GRN tasks' — comprehensively covering the model's task families with no vague filler.	5 / 5
Completeness	Clearly answers 'what' ('Embed and annotate single-cell expression data with scGPT, a foundation model for single-cell biology') and 'when' with an explicit 'Use this skill when: (1)... (2)... (3)...' trigger enumeration.	5 / 5
Trigger Term Quality	Good natural-term coverage ('single-cell', 'AnnData', 'clustering', 'integration', 'cell-type annotation', 'perturbation', 'GRN'), but common synonyms and extensions users might say ('scRNA-seq', 'transcriptomics', '.h5ad') are missing.	4 / 5
Distinctiveness Conflict Risk	Clear niche (scGPT foundation-model embeddings/annotation) with an explicit boundary — 'For probabilistic single-cell models (scVI etc.), use the scvi-tools library' — minimizing mis-trigger risk.	5 / 5
	Total	19 / 20 Passed

Validation

81%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: aipoch/open-science
Path: resources/skills/scgpt/SKILL.md
Commit: bf35648

Reviewed: 8 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.