Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/paper-pipeline
Orchestrate the complete post-first-draft polishing pipeline for an academic LaTeX paper by invoking five existing skills in fixed order: (1) paper-polish, (2) paper-self-revise, (3) paper-style, (4) paper-polish again, (5) reference-verify.
Passed 16 of 19 test cases with the skill, and 8 of 19 without it.
and 6 more in the full report
9 fixed9 passed either way4 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
After completing Stage 2 (self-revision) on `./macro_finance`, what exact filename should be used to store the snapshot of `main.tex` inside `pipeline_backups/`?
| Checked | Without skill | With skill |
|---|---|---|
| Names the stage snapshot file main_after_stage2.tex. | Missed | Passed |
The instructions your model reads when the skill loads.
This skill runs the full post-first-draft polishing process on an academic paper by invoking five existing skills in a fixed order. It is an orchestrator: it does not duplicate any sub-skill's logic. Each stage is executed by invoking the corresponding skill with the Skill tool and following that skill's own workflow to full completion before moving on.
The pipeline (fixed order — never reorder or skip without explicit user request):
| Stage | Skill | Purpose in the pipeline |
|---|---|---|
| 1 | paper-polish | Clean mechanical, consistency, and citation errors so the self-review can focus on substance |
| 2 | paper-self-revise | Deep internal review and content revision on a clean draft |
| 3 | paper-style | Restructure title + sections to the target journal, after content has stabilized |
| 4 | paper-polish | Second pass: catch issues introduced by Stages 2–3 (cross-references, merge seams, new prose) |
| 5 | reference-verify | Final reference audit, once the citation set is settled by all prior edits |
The order matters: polishing before reviewing keeps the review substantive; restructuring after content revision avoids churn; the second polish exists specifically because Stages 2–3 modify text and structure; reference verification runs last because earlier verification would be partially invalidated by later edits.
Input: A project folder containing main.tex (plus tables/figures, ref.bib).
Output: The fully polished main.tex + ref.bib, per-stage backups, a ref_verify_report.xlsx, and a final pipeline report.
Receive folder path from $ARGUMENTS or ask. Verify main.tex exists with Glob.
Dropbox conflict check: Glob the project folder for *冲突副本* and *conflicted copy*. If found, stop and ask the user to resolve which version is canonical before anything else. Re-run this check between every stage (concurrent Overleaf editing can silently revert edits, and a 5-stage pipeline is a long exposure window).
Check for a previous pipeline run: if pipeline_state.json exists in the folder and has incomplete stages, show its status and ask: resume from the first incomplete stage, or start over from Stage 1? If resuming, skip directly to that stage.
Collect the target journal now (needed by Stage 3, asked upfront so the pipeline never stalls mid-run). If given in $ARGUMENTS, use it. Otherwise use AskUserQuestion with common options (JFE / JF / RFS / Management Science) plus "Other" for free-form input.
Ask for the interaction mode with AskUserQuestion — this is the single most important setup question, because five interactive skills back-to-back can mean hundreds of approval prompts:
Create the backup baseline: make a pipeline_backups/ subfolder and copy main.tex → pipeline_backups/main_stage0.tex and ref.bib → pipeline_backups/ref_stage0.bib. After each completed stage, snapshot main.tex as main_after_stage<N>.tex. These snapshots are the rollback path if a later stage goes wrong.
test cases passed, vs 8 / 19 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.