stereo-seq-publication-story - SKILL.md Agent Skill

name: stereo-seq-publication-story description: Use whenever a Stereo-seq/STOmics/spatial transcriptomics request asks to reuse, cite, report, compare, or learn from real published Stereo-seq papers, paper templates, article-derived workflows, manuscript figure logic, publication story, scientific story, paper experience, provenance, DOI, code/data source, or "which paper/template was reused". Also use for Chinese requests mentioning 真实论文, 文章模板, 论文模板, story templates, 论文故事线, 复用论文经验, 复用了哪些论文, 原始论文 DOI, 代码/数据来源, 发育图谱, 器官发生, mouse embryo, organogenesis, developmental atlas, or when designing a paper-style analysis plan from a new Stereo-seq dataset. This skill searches bundled unified paper profiles before mapping to local Stereo-seq analysis/plotting skills.

Stereo-seq Publication Story

Use this skill to turn a new Stereo-seq analysis request into a paper-like scientific storyline grounded in previously published Stereo-seq article patterns.

Use This For

Rapidly proposing testable scientific questions from a new dataset.
Matching tissue, species, disease, developmental stage, perturbation, or analysis goal to similar Stereo-seq papers.
Building a figure-by-figure mini manuscript plan before running downstream workflows.
Deciding which existing local Stereo-seq skills should be combined, based on similar published analyses rather than fixed tool-selection rules.
Explaining which paper story patterns were reused and what must still be validated on the user's data.

Default Requirements

Search local references first. Do not browse for papers or code unless the bundled corpus is insufficient or the user explicitly asks.
Start with scripts/search_paper_stories.py using the user's tissue, species, condition, stage, biological question, analysis keywords, tool hints, or skill tags. The default search covers all bundled unified paper profiles.
When code provenance or reusable scripts are needed, also run scripts/search_code_repositories.py against the bundled article/method code registry before considering external search.
Read only the top relevant references/paper_profiles/papers/Sxxxx.md files needed for the task.
Treat the bundled material as a paper-derived evidence base, not as proof about the user's dataset. Separate:
- what the reference paper showed;
- what the user's dataset has already demonstrated;
- what remains a hypothesis to test.
Do not hard-code tool selection. Infer candidate workflows from similar papers, observed tools, story archetypes, and available local Stereo-seq skills.
When recommending downstream work, name the relevant local skills and the paper evidence that motivated each choice.
When writing a final report, include paper id, title, DOI, original code/data source if present in the digest, and whether a story template or digest was reused.

Workflow

Parse the user's dataset and goal: species, tissue, stage, condition, spatial unit, annotations, matched sc/snRNA data, and expected biological question.
Search the local paper-story corpus:
- python scripts/search_paper_stories.py --query "mouse brain cortex layer spatial domain cell type" --top 8
- python scripts/search_paper_stories.py --species mouse --tissue brain --skill-tags spatial_domain cell_type_mapping spatial_programs --top 8
Search local code-source evidence when executable or figure-template provenance is needed:
- python scripts/search_code_repositories.py --query "cellbin segmentation histology boundary Stereo-seq" --top 12
- python scripts/search_code_repositories.py --skill stereo-seq-spatial-domain-discovery --query "mouse embryo domain marker" --top 12
Read the most relevant unified paper profiles first. Use references/github_code_registry.tsv and per-skill references/code_candidates.tsv only as local source-code navigation, not as a claim that every file was executed.
Extract reusable story components:
- central question and biological gap;
- figure order template;
- minimum evidence chain;
- analysis modules used by the reference paper;
- validation requirements and common failure modes.
Convert the pattern into a dataset-specific plan:
- candidate scientific questions;
- analysis modules to run;
- figure storyboard;
- expected outputs and decision points;
- risks, missing metadata, and unsupported claims.
Map planned analysis modules to available local skills. Common pairings include:
- end-to-end workflow routing: stereo-seq-analysis-workflow
- project metadata, multi-sample QC, batch/condition handoff: stereo-seq-project-orchestration
- QC/setup: stereo-seq-quality-control-preprocessing
- image/expression registration: stereo-seq-image-registration
- cellbin/segmentation/histology: stereo-seq-cellbin-segmentation
- annotation/label transfer: stereo-seq-cell-type-mapping
- domain/layer discovery: stereo-seq-spatial-domain-discovery
- marker/pathway/program analysis: stereo-seq-spatial-programs
- replicate-aware condition inference: stereo-seq-statistical-design
- spatial interpretation: stereo-seq-spatial-cell-type-interpretation
- CCI: stereo-seq-spatial-communication or stereo-seq-cell-cell-interaction
- GRN/regulon: stereo-seq-spatial-grn-regulon
- trajectory/time/state transition: stereo-seq-developmental-trajectory
- 3D/serial sections: stereo-seq-3d-reconstruction
- manuscript figures: stereo-seq-publication-plotting
Report the provenance of the reused story patterns and code-source templates, then explain how they were adapted to the user's data.

Reference Layout

references/paper_profile_all_index.tsv: searchable index of all bundled unified paper profiles with tissue, tools, skill tags, story/reuse fields, and profile paths.
references/paper_profiles/papers/Sxxxx.md: one unified profile per paper, combining objective corpus evidence and story/reuse cues.
references/story_template_schema.md: schema used to interpret story templates.
references/github_code_registry.tsv: curated local index of article-owned or method-with-example GitHub/code repositories linked to paper ids, DOI, reusable files, and relevant local skills.
scripts/search_code_repositories.py: searchable interface over the code registry.

Output Expectations

For a story-design or analysis-planning request, produce:

2-5 candidate scientific questions that match the user's dataset.
A recommended primary story archetype and 1-3 fallback archetypes.
A concise figure storyboard with required analysis modules.
Matching reference papers with paper id, title, DOI, and what was reused.
A mapping from each module to existing local Stereo-seq skills.
Caveats describing missing metadata, weak evidence, unsupported claims, or package/environment blockers.

For a completed analysis report, include:

The analyses performed and files generated.
Which paper-derived profiles were reused.
Original paper DOI and code/data source as available in the digest.
What was changed to adapt the template to the user's dataset.