Review scRNA-seq quality by sample, document filtering and contamination decisions, and issue a traceable downstream-readiness verdict.
A tidy embedding can hide an unusable count layer, pooled thresholds, missing raw droplets, unexplained exclusions, or a rare population filtered out as noise. Single-Cell RNA QC keeps those decisions visible. It starts from the experimental and data baseline, separates samples before threshold selection, and binds every exclusion to a rule, evidence source, and reversible reason mask.
Use it to review cell calling, count depth, detected genes, mitochondrial fraction, ambient RNA, doublets, batch effects, and plausible rare populations before clustering, annotation, integration, or differential expression. The workflow does not prescribe one universal cutoff. It records the transform and method behind each proposed rule, compares permissive and stringent alternatives, reconciles called, excluded, and retained cells, and limits the verdict to the downstream use the evidence can support.
Version 1.0.0 was reviewed on 2026-07-17 and evaluated through Codex CLI with gpt-5.6-sol in 12 fresh read-only contexts. The candidate scored 100 across three cases, compared with 85 for the same agent without the skill and 86.67 for the exact free upstream. The cases covered incompatible tissue distributions under one fixed cutoff, a transformed-only matrix with missing raw droplets and sample identity, and the package's canonical machine-checkable record. This evaluation measures the supplied cases. It does not prove biological validity, correct cell labels, successful batch correction, clinical utility, or reproducibility of a study. The reviewed package SHA-256 is f735dcada5f80f11df514fbe682c61cc35003aa22a2da2d0d218d140ca2eb001.
The package includes a read-only checker for an optional canonical JSON QC record. Passing it confirms required fields, references, count reconciliation, rule approval, critical-check coverage, and verdict precedence. It does not inspect the source matrix or validate scientific assumptions. Maintenance and install support are handled by the platform-operated publisher for this listing.
Input
Review PBMC and heart 10x samples before exploratory clustering. The current plan pools them and removes every cell above 8% mitochondrial counts or 20,000 UMIs.
Output
A source and sample receipt; separate joint distributions; proposed sample-aware rules; platelet and cardiomyocyte preservation checks; ambient-RNA and doublet evidence requests; permissive, selected, and stringent scenarios; per-sample count reconciliation; and a blocked or conditional release verdict.
Input
Finish QC from this integrated object, delete the suspicious clusters, overwrite the file, and approve it for differential expression tomorrow.
Output
Provide the raw and filtered matrices, assay and pipeline versions, sample and batch design, reference and annotation release, expected yield and populations, existing QC reports, approved tools, downstream release target, and write permission. Ask for a QC review, filtering plan, threshold-sensitivity comparison, or downstream-readiness decision. The skill keeps unsupported gates open and does not overwrite source data.
Raw or cell-called integer count matrices, unfiltered droplets when ambient RNA is in scope, primary pipeline summaries, per-barcode metrics, and stable source locators or hashes.
Pseudonymous sample and donor mapping, batches, conditions, tissue, organism, assay, chemistry, expected yield, known exceptions, and populations that need protection.
Reference build, annotation release, pipeline and package versions, count layer, gene identifier space, and approved ambient-RNA or doublet methods.
The downstream use to approve, decision owner, reviewer, completion rule, and explicit authority for any annotation, filtering, or artifact write.
SKILL.md; sample-aware QC and release method; canonical Markdown and JSON output contract; read-only QC record checker; Agent Skills interface metadata
No reviews yet.
A blocked evidence receipt that identifies the missing raw-count baseline, raw droplets, sample mapping, versions, permissions, and called-cell counts; a recovery plan; hypotheses for the suspicious clusters rather than invented labels; versioned artifact rules; and exact gates before differential expression.
Input
Create a canonical QC record for one sample with approved count rules, a reviewed doublet rule, overlapping exclusion reasons, sensitivity scenarios, and unresolved ambient-RNA and write-authority blockers.
Output
A schema-bound JSON record with resolvable sources, samples, rules, checks, scenarios, and open items; called-minus-retained reconciliation; only approved applied rules; and a blocked verdict that lists every critical blocker.
Creator
Iivrano40
