| name | sci-using-kit |
| description | Use when starting or resuming any SCI paper task in this workspace, including manuscript revision, citation repair, result writing, review preflight, LaTeX editing, arXiv packaging, or multi-agent handoff |
SCI Using Kit
Overview
This is the entry skill for the SCI using kit skill family. It routes the task to the right evidence-bounded branch skill before any manuscript edits.
Core principle: choose the paper task type first, then act with the smallest evidence-safe change.
Startup Rule
Before editing or advising, read the project state:
skills/sci-using-kit/references/prompt-kernels.cn.md- Project files if present:
PROJECT_BRIEF.md, CLAIMS.md, EVIDENCE_MATRIX.md, DECISIONS.md
- Existing manuscript files, figures, tables, logs, bibliography, and public-release constraints
On Windows, use explicit UTF-8 for Chinese scaffold files:
- PowerShell reads:
Get-Content -Encoding UTF8
- Python tools: set
$env:PYTHONUTF8='1' before validation or scripts
- Avoid relying on the system GBK default for
.md, .tex, .bib, or .yaml
Project Phase
Classify the project phase before routing:
| Phase | Meaning | Typical next branch |
|---|
| kickoff | new paper/project, unclear dossier | sci-issue-intake |
| drafting | building sections from evidence | sci-paper-revision, sci-evidence-discussion |
| revision | advisor/reviewer feedback, major rewrite | sci-issue-intake, then task branch |
| arXiv preflight | public preprint packaging/readiness | sci-reviewer-preflight, sci-submission-preflight |
| SCI Track B | journal-strengthening work after arXiv | sci-comparative-positioning, sci-evidence-discussion |
| repo release | public code/data/README/source release | sci-submission-preflight |
Branch Selection
| Situation | Required branch |
|---|
| New issue, reviewer comment, unclear user request | sci-issue-intake |
| Literature search, citation repair, related work | sci-literature-vetting or literature-review-search |
| Section rewrite, de-AI prose, claim tightening | sci-paper-revision |
| Tables/logs/results need discussion text | sci-evidence-discussion |
| Figure planning, image-generation prompts, graphical abstracts, workflow diagrams, evidence snapshots, captions, or Overleaf figure assets | sci-figure-generation |
| Claims of faster/better/improved/efficient/novel vs prior work | sci-comparative-positioning |
| Final arXiv/SCI readiness review | sci-reviewer-preflight |
Live .tex, figure/table, citation, author, package edits | sci-latex-editing |
| arXiv source package or submission checklist | sci-submission-preflight |
Mandatory Gates
- If a sentence includes improvement over prior work, run
sci-comparative-positioning.
- If a paragraph uses a result number, trace it to a table, log, script, or confirmed fact.
- If a figure uses a number, unit, axis, or visual comparison, trace it to evidence and check that the visual encoding is not misleading.
- If a citation supports a claim, verify the source before using it.
- If evidence is missing, write a gap note, not a stronger claim.
- Preserve the prompt kernels: SOP routing must not remove claim/evidence discipline, literature context isolation, reviewer perspective, or section-scoped editing.
Output Contract
Every task response must state:
- task type and branch used
- evidence consulted
- files changed or recommended changes
- residual risks
- verification performed or not performed
Use this fixed block near the start of the response:
**Routing Decision**
- Project phase:
- Task type:
- Primary branch skill:
- Additional gate skills:
- Evidence bundle to read:
- Output expected:
Red Flags
Stop and reroute if you are about to:
- rewrite the whole paper without a scoped brief
- turn HIL/simulation into flight or RF validation
- compare against prior work with different task, data, hardware, or metric
- hide limitations to make prose sound stronger
- copy marketing claims back into the manuscript
See references/skill-family-map.md for the intended workflow map.
